Compare commits

..
Author SHA1 Message Date
jason.woltje b7239bb66e agents: row 195 Case 1 install design r6 (#1693): Duckworth r5 B1 (confirm's result lines carry the activation's op, from, to and the rollback set's BACKUP, so from_log reads .53's log; sequence 7), B2 (MIGRATING recorded after backup_store and before migrate-schema.mjs, as TAKEN today; sequence 8), N1 marker step makes $RS, N2 gate --since, N3 apparmor.d word grep baseline, N4 item 4 under systemd-run with NoNewPrivileges, N5 plan --expect-from OLD 2026-10-20 16:53:53 -05:00
jason.woltje 0b6d29e909 agents: row 195 Case 1 install design r5 (#1693): Duckworth r4 B1 (one result line per activation, FAILED file and --cause, marker written first, refusals log no activation, real-sequence tests), B2 per mos-1098/mos-1101 (case1-pins.sh and case1-hostcheck.sh, seven-item drift stop with apt hook and boot unit, ops-01 owner, bwrap closed on drift), N1-N4; Console on 127.0.0.1:732 with CAP_NET_BIND_SERVICE, tester 7333 as P10; chains on row 190 r3 and row 230 (C2-13a), node under /opt, four factory profile items for Gosalyn 2026-10-20 16:46:21 -05:00
jason.woltje 0075be3581 agents: row 195 Case 1 install design r4 (#1693): Duckworth B1 (root holds only the go's values and the unit files; confirm and revert find the one open activation, AWAITING-UNITS; rollback --to checked against from_log previous; A2 failure path; root step log), B2 (confirm runs gate with --timeout; .53 runs the PM; external PM check reads launches list with ActiveEnterTimestamp, no journal; pm-on refuses), N1-N5; mos-1096 bwrap-userns-restrict from the pinned deb, dpkg-statoverride root:mosaic 0750, P9 accepted by Mos; gosalyn-0046 two units and args file; managed settings for DISABLE_AUTOUPDATER; sage-1481 C2-13a 2026-10-20 16:08:53 -05:00
jason.woltje 9a16d7795d agents: row 195 Case 1 install design r3 (#1693): Duckworth B1, one ordered .53 activate with a runner per step (case1-repin.sh, stack-release --units external with confirm and revert), stopped-store backup under $DATA/backups/stack; N1-N6; mos-1091 4a-4c with the bwrap AppArmor profile; gosalyn-0034 C2-13 items 2026-10-20 15:56:54 -05:00
jason.woltje 818b79696d agents: row 195 Case 1 install design r2 (#1693, brief blob 45be7f09af, Sage rulings sage-1451) 2026-10-20 15:38:55 -05:00
jason.woltje 088087ac96 queue: revs 2248-2361+. Sage: rows 210 (#424 host access, Gizmoduck) and 211 (#426 force-push guard, Rocko) added and briefed; row 195 re-briefed to the Case 1 install on .53; review requests for 200 r1/r2 and 202; plus earlier Sage revs. Other seats' revs: Launchpad 2254, 2283-2284, 2286, 2311-2313, 2328-2329, 2344, 2347-2348 (rows 192/200); Honker 2255; Morgana 2257, 2320, 2326-2327 (row 204); Gander 2262; Herb 2265; Dewey 2268; Bushroot 2273; Megavolt 2276; Webby 2303, 2316, 2339, 2355 (row 192/200 verdicts); Gizmoduck 2304, 2318-2319 and the row 210 claim (rows 201/210); Quackerjack 2317, 2341-2342 (row 202); Duckworth 2323, 2335, 2353 (rows 201/204/202 verdicts); Darkwing 2354, 2357 (row 190 in-review) 2026-10-20 15:34:55 -05:00
jason.woltje a0fbefd48d plans: pre-push force-push hook and failing busy check brief (Vikunja #426, mos-1078, mos-1083; owner Rocko, reviewer Dewey) 2026-10-20 15:31:17 -05:00
jason.woltje f24f2cd40d plans: row 195 brief becomes the Case 1 install on 10.1.1.53 (Q181 hosts, Q182/Q183 recorded acceptance, Q185 existing account, chained on rows 190 and C2-13; ops-01 root steps under #423 step 1; sage-1439) 2026-10-20 15:24:33 -05:00
jason.woltje 9ba31b5b3f agents: Honker's PRD 1.7 draft r2 on refactor for the prd.approve raise (11cb0423b, sha256 e896746df595d1b5155130ca163fbf63b0dc38a87a8dd154af77f1a6d1328647, Duckworth approved duckworth-0039, Vikunja #420) 2026-10-20 15:20:37 -05:00
jason.woltje 9db578ff54 plans: agent host access with sudo through the credential store brief (#1707, Vikunja #424, Q183/Q184, chained on row 199) 2026-10-20 15:09:59 -05:00
jason.woltje 8eb9ddd12c plans: FED R8 brief to r4.11 (b93f098f0, recorded acceptance, Duckworth N1/N2, #416 date), R5 context source (sage-1419), 193a makes covers level required (duckworth-0033 N1) 2026-10-20 15:06:22 -05:00
jason.woltje ae3c36f248 agents: Gizmoduck seat dir and roster row (FED R1 owner, #414) 2026-10-20 14:42:15 -05:00
jason.woltje a2af7ab495 plans: FED R1, R5, R2 and R8 briefs (#414, r4.9 a1daa3c07) and row 193a-d build briefs (#1691); Quackerjack seat dir and roster row 2026-10-20 14:41:03 -05:00
51 changed files with 5805 additions and 3659 deletions
+2
View File
@@ -23,6 +23,8 @@ not fleet seats. Keep changes in `/mnt/storage/src/mosaic-stack`; do not modify
| Herb | Builder, a second gate lane (#1695) | T3 session (Codex, gpt-6.1-sol), started by Sage under mos-0977 | none, T3 only |
| Bushroot | Builder, retire the control board writer on 7331 (#1696) | T3 session (Claude, claude-haiku-5-5), started by Sage under mos-1008 | none, T3 only |
| Megavolt | Builder, MVP T1 item 6 (agent credentials and git identity, #1697) | T3 session (Claude, Opus 5.5), started by Sage under mos-0974/0977 | none, T3 only |
| Quackerjack | Builder, federation R5 and R2 (#414) | T3 session (Claude, claude-haiku-5-5), started by Sage under mos-1023 | none, T3 only |
| Gizmoduck | Federation design (#414) and builder of FED R1 | T3 session (Claude, Opus 5.5, thread 9312a896) | none, T3 only |
| Webby | Reviewer: Case 2 designs, rows 192 and 194 | T3 session (Claude, Opus 5.5, thread 03086006), started by Mos (mos-0981) | none, T3 only |
| Duckworth | Reviewer: federation designs, row 193 | T3 session (Claude, Opus 5.5, thread ffceafec), started by Mos (mos-0977) | none, T3 only |
| Sage | Project lead: coordination, review, integration; earlier DYOR strategy records retained | T3 session (Claude Code); Pi launcher `zai/glm-5.3:high` retained | `agents/sage/launch.sh` |
+872
View File
@@ -0,0 +1,872 @@
# Row 195 design: the Case 1 install on 10.1.1.53 (r6)
Darkwing, 2026-10-20. Brief: `docs/plans/2026-10-20_hub-own-uid.md`, blob 45be7f09afc1 (f24f2cd40,
sage-1439). Gitea #1693. Reviewer Duckworth, gate owner Sage. Chained on row 190's candidate (round 3,
df65a3d4c on `darkwing/row190-r3`, rebased on row 192; the config canon) and on row 230's candidate
(C2-13a, #1732, Gosalyn: `scripts/case2-unit.sh` and `case2-precheck.sh`), Q177, sage-1499. Row 230's
brief is `docs/plans/2026-10-20_case2-rows.md` "C2-13a" at 5b8def2f1, from Gosalyn's design r11
(eac92d956 on `gosalyn/case2-design`, approved webby-0038). It has no candidate yet, so I read that
brief as the interface and expect a rebase. Section 2 lists what Case 1 needs from it that the brief
doesn't give.
Changes from r5 (Duckworth's REVIEW-r4.md at 62200362b on `duckworth/reviews`, duckworth-0054). This
round changes only B1's and B2's lines, their section 7 cases and the notes:
- B1. The result line `confirm` writes said op `confirm`, because `log_line` (:58) writes `$OP` and
`OP="$1"` (:1493). `from_log` (:169) reads only `adopt`, `activate` and `rollback`, so it would have
skipped every .53 activation: after adopt C0 and re-pins to C1 and C2, `from_log previous` returns
C0. Now A5's `ok` line and R4's line carry the op, from and to of the activation's `started` line,
and the backup that A3 records in the rollback set. Real sequence 7 and a mutant that writes op
`confirm`. Section 4.
- B2. r5 recorded the migration after the migrate, while today's run sets TAKEN before
`migrate-schema.mjs` (:1004-1005). A migrate that changed the store and then exited 1 would have
left no record, so R2 would skip the restore and R4 would gate OLD on a half-migrated store. A3 now
writes `MIGRATING` after `backup_store` returns and before the migrate. Real sequence 8 and a mutant
that records after the migrate. Section 4.
- N1. The marker step makes `$RS` with capture's exclusive `mkdir` (:416), and capture under
`--units external` makes none. N2. `gate` takes `--since T`, root's clock before its A2f start, as
its start floor. N3. Section 0 now has a read-only grep of `/etc/apparmor.d` for the word `bwrap`,
and item 3 matches the word against a list of exactly one file. N4. Item 4 runs under
`systemd-run` with the units' `NoNewPrivileges=yes`, as 2a' does. N5. A1's plan takes
`--expect-from OLD` and refuses `pointer-differs`.
- Not in this round: claude's path. gosalyn-0058 proposes that `--claude`, like `--node`, takes only a
root-owned path outside the home, and `/srv/mosaic/.local/bin/claude` fails that. I asked Sage to
rule and recommended moving claude under `/opt/mosaic-stack`. That change gets its own round.
Changes from r4 (Duckworth's REVIEW-r3.md at 9f4d5556b on `duckworth/reviews`; mos-1098; mos-1101;
gosalyn-0049):
- B1. An activation now has exactly one result line, written at the end. A5's gate failure and A3's
failure after its capture write no log line. Each writes a `FAILED` file next to `AWAITING-UNITS` and
exits 1, so the activation stays open for R2 and R4, and R4 writes the result. A3 writes
`AWAITING-UNITS` right after `started`. Refusals by `confirm`, `revert` and `gate` log no activation
id, so a refused `confirm` can't close one either. There are real-sequence tests on one log, and
mutants that put back r4's interim line, r4's marker order and a refusal that names the activation.
Section 4 and section 7.
- B2. The bwrap profile has an owner and a stop. `scripts/case1-hostcheck.sh` checks the profile file
against the pin in `scripts/case1-pins.sh`, the two loaded profiles, the live label of a process
bwrap runs, the statoverride and mode, that no other profile names bwrap, two sysctls and the port
floor. It has three callers:
1. A1 of every re-pin refuses `profile-drift`.
2. A root apt hook runs it after every package run, unattended ones included.
3. A boot unit runs it before the bus starts.
On drift, callers 2 and 3 unload the profile and close `/usr/bin/bwrap` to `mosaic`, so every role
session refuses to start. They write `DRIFT` and log it, and only root clears it. ops-01 owns the host
side under a Mos go: it reads the log in each STATE report and, before each re-pin window, compares
the upstream profile to the pin. A changed upstream profile becomes a new pin, which is its own review
round. Section 1 has the details. The facts behind it are in section 0: unattended-upgrades runs daily
on .53 with the security pocket, and the pinned bubblewrap is in that pocket.
- N1 and mos-1098 point 2. P9 has a seventh item: no `mosaic-bwrap` profile is loaded, and no profile
other than Ubuntu's attaches to `/usr/bin/bwrap`. r4 never installed `mosaic-bwrap`. F5 now compares
exact strings. P9 reads `accepted mos-1096` only while all seven items hold.
- N2: a role session gets no capabilities, so a nested bwrap fails. Section 1. N3: A3 refuses "before
it logs `started`". N4: the stale-start rule and its fixtures handle `ActiveEnterTimestamp`'s whole
seconds.
- The port squat (mos-1101 item 5, gosalyn-0049). It applies: role sessions share the host's network
namespace (`packages/seat/src/sandbox.mjs:67` unshares no net), so while the Console is down a
session can bind its loopback port. The Console moves to 127.0.0.1:732 with
`CAP_NET_BIND_SERVICE` on its unit (row 230's line), so no session can take it. The tester port stays
7333, because serve refuses a tester port below 1024. That residue is P10, accepted under Mos's
reading. Case 1 can't meet the preview trio or the proof pair, because .53 has no proxy. Section 2
names the factory serve set for row 230.
- Jason's Q9 (Vikunja #429) is still open. If install and upgrade become product commands, these
steps become that installer's configuration and row 195 rebases on its row (sage-1491). Section 6.
## 0. What I checked on the host
Read only, as `localadmin` over ssh, 2026-10-20 between 20:40Z and 21:00Z, and again at 21:27Z for the
last seven rows (the brief allows this):
| Item | 10.1.1.53 today | Needed |
|---|---|---|
| OS | Ubuntu 24.04.5, kernel 6.8.0-124, x86_64, 12 cores, 94 GiB, 217 GiB free on `/` | fits |
| `mosaic` user, group, `/srv/mosaic` | absent | step 2a creates them |
| node | `/usr/local/bin/node` v22.22.2, root-owned | 24 or later (every `packages/*/package.json` engines line); dragon-lin runs 26.8.1 |
| bwrap | absent, no statoverride; apt candidate `bubblewrap 0.9.0-1ubuntu0.3` | every role session runs under `/usr/bin/bwrap` (`packages/seat/src/sandbox.mjs:43`) |
| user namespaces | `kernel.apparmor_restrict_unprivileged_userns = 1`, apparmor 4.0.1, no bwrap profile loaded | bwrap needs a user namespace, which this setting refuses unless a profile allows it |
| `apparmor-profiles` | not installed; candidate `4.0.1really4.0.1-0ubuntu0.24.04.9`, whose `/usr/share/apparmor/extra-profiles/bwrap-userns-restrict` has sha256 `11d39094f044f0cda0febb3ad517b830301da6b2ce929664af09ee9e4dd264f9` (read from the deb on dragon-lin) | section 1 |
| `fs.protected_hardlinks`, `fs.protected_symlinks` | 1, 1 | mos-1096 item 2 |
| journald | `SplitMode=uid` (the default) | section 4, B2 |
| `useradd` | `USERGROUPS_ENAB yes` | 2a passes `--user-group` anyway |
| claude | not on localadmin's PATH | the factory's harness; dragon-lin runs 2.1.304 |
| ports 7330-7333 | free (no listener) | tester |
| `localadmin` | NOPASSWD sudo (#423 step 1, applied 20:05:47Z, mos-1066, mos-1077) | the root steps run through it |
| `jarvis` | groups sudo, docker, lxd, adm | #423 step 4 removes all four (section 5, P5) |
| unattended-upgrades | on: `APT::Periodic::Unattended-Upgrade "1"`, `apt-daily-upgrade.timer` enabled. Allowed origins `${distro_id}:${distro_codename}`, `-security`, ESM apps-security and ESM infra-security. No `Package-Blacklist` | B2: packages change daily without a window |
| bubblewrap's pocket | `0.9.0-1ubuntu0.3` is in both `noble-updates` and `noble-security`, so its later security updates install unattended | B2 |
| apparmor | installed `4.0.1really4.0.1-0ubuntu0.24.04.8`. The `.9` candidate is in `noble-updates` only, which unattended-upgrades doesn't take. `aa-status --json` works | B2 |
| profiles that name bwrap | at 21:51:46Z `grep -rlw bwrap /etc/apparmor.d` and `grep -rlF /usr/bin/bwrap /etc/apparmor.d` both find nothing (exit 1). After 2a' the word grep lists only `bwrap-userns-restrict` | N1, P9 item 7, hostcheck item 3 |
| `net.ipv4.ip_unprivileged_port_start` | 1024 | the port squat |
| listeners below 1024 | 22 (sshd) and 53 (the resolver) only; 732 is free | the Console's port (section 1) |
| other listeners | 4003, 6052, 6055, 8123, 8765, 14242, 18555, and on loopback 1933, 5905, 6085, 6379, 6380, 9119, 9225, 18554, 18789, 38147 | none in 7330-7333 |
## 1. The install as inputs to the Case 2 scripts
Instance. Row 190's canon with the default name, `dev`, as Case 2 uses it (Gosalyn design 9). `~` is
`mosaic`'s home, `/srv/mosaic`, so:
| Input | Value |
|---|---|
| user | `mosaic`, primary group `mosaic` with no other member |
| home | `/srv/mosaic`, 0700, owner `mosaic` |
| config dir | `/srv/mosaic/.config/mosaic-dev` |
| data root | `/srv/mosaic/.mosaic-dev`, marker `instance.json` `{"instanceVersion":1,"name":"dev"}` |
| business | `mosaic-stack`, file `<configDir>/businesses/mosaic-stack.json` (section 2) |
| profile | `--profile factory` (row 230, C2-13b) |
| roles | the factory set in section 2 |
| human id | Jason's id, given in the window's go. Not `tester` (`broker.mjs:245`) |
| model credential | `<configDir>/secrets/claude-session.env`, 0600 (row 190's `SESSION_ENV` for the default instance) |
| token slot path | none for this row (section 2) |
| Console port | 127.0.0.1:732, below the port floor, bound through the Console unit's `CAP_NET_BIND_SERVICE` (row 230). The port squat below says why |
| tester port | 127.0.0.1:7333 (P10) |
| board | none. UNITS has no board unit, and the Console gets no `--board` |
| harness | claude at `/srv/mosaic/.local/bin/claude`, pinned through `MOSAIC_CLAUDE_BIN` (decision 91), which the bus unit's `@CLAUDE_ENV@` line sets (section 2) |
| node | `/opt/mosaic-stack/node-<version>/bin/node`, a pinned 26.x release, `root:root`. Row 230 refuses a `--node` under the home or not owned by root |
| PM | on: the bus unit's `@PM_FLAGS@` renders ` --pm` (section 2, section 4) |
| source | `/srv/mosaic/src/mosaic-stack` (`mosaic`'s checkout, section 4) and `/root/mosaic-stack` (root's) |
Why `dev` and not a named instance: one instance per OS user, and Case 2's rows already assume
`/srv/mosaic/.config/mosaic-dev`. A second instance under `mosaic` on .53 would need a name, and row 190
makes that a flag, not a move.
Why node lives under `/opt/mosaic-stack`: changing `/usr/local/bin/node` would change what Hermes and
anything else on the host runs, and Q181 keeps Hermes running. r4 put node under `/srv/mosaic`, but row
230 requires a root-owned `--node` outside the home. Nothing else on the host puts `/opt/mosaic-stack`
on its PATH.
PATH for `mosaic`'s steps. `sudo -u mosaic -H` resets PATH to sudo's `secure_path`, which finds
`/usr/local/bin/node`, v22. The precheck would then refuse, and `scripts/mosaic` would run v22 (it runs
`exec node` from PATH, `scripts/mosaic:22-40`). So every `mosaic` step in the run sheet and in
`case1-repin.sh` runs as `sudo -u mosaic -H env PATH=<node dir>:/usr/local/bin:/usr/bin:/bin ...`, the
list row 230 renders as the units' `@PATH@`. `case1-repin.sh` builds it from the `--node` in
`case1-unit.args`.
The units (row 230, design r11). Two system units, from the existing templates, in start order:
`[email protected]` from `packages/cli/systemd/mosaic-bus.service.in`, then
`mosaic-webui.service` from `packages/webui/systemd/mosaic-webui.service.in`. Root stops them in the
reverse order. `@REPO@` is the copy dir, and `%h` becomes the literal home. Row 230 adds
`ProtectSystem=strict`, `ReadWritePaths=/srv/mosaic`, `ProtectHome=yes` and `PrivateTmp=yes` to both, and
`AmbientCapabilities=CAP_NET_BIND_SERVICE` with `CapabilityBoundingSet=CAP_NET_BIND_SERVICE` to the
Console only. The Console unit has no `[Install]`, so root starts it by hand after a reboot until Jason
rules (mos-1097). The install call is:
```
case2-unit.sh --profile factory --user mosaic --home /srv/mosaic --business mosaic-stack --copy ID \
--node /opt/mosaic-stack/node-<version>/bin/node --pm --claude /srv/mosaic/.local/bin/claude \
-- --port 732 --business mosaic-stack --tester-port 7333
```
`--pm` and `--claude` are the two inputs row 230's brief doesn't have yet (section 2). A re-install
must pass the same `--node`, flags and serve flags, and only the copy id may change (row 230's
re-install check). So 2c writes the whole argument list, without `--copy`, to
`/etc/mosaic-stack/case1-unit.args`, `root:root` 0644, one argument per line. Every later
`case2-unit.sh` call in section 4 reads that file and adds `--copy`. `mosaic` can't write it. Changing
a serve flag is a re-install with its own go, not a re-pin.
The port squat (mos-1101 item 5). Role sessions share the host's network namespace:
`packages/seat/src/sandbox.mjs:67` unshares the user, PID and IPC namespaces and not the network. So
while the Console is down, which A2 to A4 of every re-pin and any crash before `Restart=on-failure`
brings it back both cause, a session can bind 127.0.0.1:7332. It would then receive Jason's `ssh -L`
traffic, and it would answer the gate's `/healthz` (`stack-release.sh:485-487`) in the Console's place.
At 732 that can't happen. The floor is 1024 (section 0), a session has no capabilities (below), and
`mosaic`'s own processes outside the sandbox have none either. Only the Console unit holds
`CAP_NET_BIND_SERVICE`. Its children, the human-cli and read scripts it spawns
(`packages/webui/src/bus.mjs:31`, `reads.mjs:79`), inherit the ambient capability. No role session
descends from the Console: sessions are children of the bus unit, which has no capability line. The
gate needs no change, because `console_port` (`stack-release.sh:213-215`) reads the Console's `--port`
from the pointer. Row 230's precheck refuses when the floor is at or below the Console's port, and
`case1-hostcheck.sh` checks it after every package run (B2). The tester port can't go below 1024,
because serve refuses that (`packages/webui/src/cli.mjs:33`), so P10 records it.
Node download (N4). The run sheet pins one release, its tarball name and its sha256. As root, in a
root-owned temp dir: fetch the tarball, `SHASUMS256.txt` and `SHASUMS256.txt.asc`; `gpg --verify` the
`.asc` against the Node release keys listed in the nodejs/release-keys repository, imported into a
keyring under `/etc/mosaic-stack/node-keys`; check that the tarball's line in the verified file is the
pinned sha256; then `sha256sum -c` the tarball and unpack it to `/opt/mosaic-stack/node-<version>`,
`root:root`, no file writable by group or other. Any mismatch stops the step. Rollback: `rm -r` that
dir.
Claude (N5). The run sheet pins the version, 2.1.304 unless the window's plan names a newer one. As
`mosaic`: `npm install --prefix /srv/mosaic/.local @anthropic-ai/claude-code@<pin>`, with `@<pin>`
exact. `/srv/mosaic/.local/bin/claude` is the path `MOSAIC_CLAUDE_BIN` names. The autoupdater stays off
through `/etc/claude-code/managed-settings.json`, `root:root` 0644, holding
`{"env":{"DISABLE_AUTOUPDATER":"1"}}`. Root writes it in 2a'. Why there:
- `ENV_ALLOW` (`packages/seat/src/session.mjs:83-88`) doesn't pass `DISABLE_AUTOUPDATER`, so a unit
`Environment=` line never reaches a role session (gosalyn-0046 item 3).
- A role session's HOME is the run's own `home/`, and the sandbox doesn't show `~/.claude`
(`session.mjs:92-93`, `sandbox.mjs:16`). So `mosaic`'s claude settings file doesn't reach one either.
- The sandbox mounts `/etc` read-only (`sandbox.mjs:85`), and Claude Code reads managed settings from
`/etc/claude-code/` whatever HOME is. `mosaic`'s own CLI runs, like the precheck's marker prompt
and F6, read it too.
The pinned binary is a read-only bind inside the sandbox (`sandbox.mjs:9-10`), so a session can't
rewrite it anyway. The managed file covers `mosaic`'s runs outside the sandbox. If the installed
version doesn't honour `env` there, F6 catches the drift. Rollback: `rm` the file.
bwrap and the user namespace (mos-1091 4a, 4b; mos-1096 item 2; N1). Three root steps in 2a', in this
order, each with its rollback. The `mosaic` group must exist first (2a).
1. `apt-get install bubblewrap=0.9.0-1ubuntu0.3`, the version the run sheet pins. Then record
`stat -c '%U:%G %a' /usr/bin/bwrap` to `/etc/mosaic-stack/bwrap.orig` (I expect `root:root 755`).
Rollback: `apt-get remove bubblewrap`.
2. Ubuntu's own `bwrap-userns-restrict`, one file, not the package. `apt-get download
apparmor-profiles=4.0.1really4.0.1-0ubuntu0.24.04.9` into a root-owned temp dir, extract
`./usr/share/apparmor/extra-profiles/bwrap-userns-restrict` with `dpkg-deb --fsys-tarfile`, check
its sha256 is `11d39094f044f0cda0febb3ad517b830301da6b2ce929664af09ee9e4dd264f9`, and install it as
`/etc/apparmor.d/bwrap-userns-restrict`, `root:root` 0644. Load it with `apparmor_parser -r
/etc/apparmor.d/bwrap-userns-restrict`. `aa-status` then lists `bwrap` and `unpriv_bwrap` in
enforce mode. Why one file: the package also installs 20 profiles under `/etc/apparmor.d`
(`usr.sbin.dnsmasq`, `bin.ping`, `usr.sbin.smbd` and more), which would load at the next boot and
could change services on the host that have nothing to do with Mosaic.
What the profile does: `bwrap` gets `userns`, `capability` and `mount` for its own setup, and
every program it runs is stacked under `unpriv_bwrap`, which has the same file and network rules
with `audit deny capability`. So a role session gets a user namespace and no capabilities in it.
The sysctl stays at 1, so nothing else on the host gains user namespaces.
One consequence (N2): a bwrap that a session starts inside its own sandbox fails, because it needs
capabilities in the namespace. Claude Code's own sandbox is such a bwrap, and it's off by default,
so a role session doesn't use it. F3 shows a session still answers under the profile.
Rollback: `apparmor_parser -R /etc/apparmor.d/bwrap-userns-restrict`, then `rm` it.
3. `dpkg-statoverride --update --add root mosaic 0750 /usr/bin/bwrap` (Mos). Only root and the
`mosaic` group can read or run bwrap, and package upgrades keep the mode. Rollback:
`dpkg-statoverride --remove /usr/bin/bwrap`, then `chown` and `chmod` to the values in
`bwrap.orig`.
4. The drift stop (B2, below): `case1-hostcheck.sh` and `case1-pins.sh` from `/root/mosaic-stack` to
`/usr/local/lib/mosaic-stack/`, `root:root` 0755 and 0644; the apt hook
`/etc/apt/apt.conf.d/80mosaic-hostcheck`; the boot unit `mosaic-hostcheck.service`, enabled. Then
`case1-hostcheck.sh --check` exits 0. Rollback: disable and remove the unit, `rm` the hook file and
the dir.
Full rollback runs 4, 3, 2, 1. Tests after 2a' and again in section 5: as `mosaic`, `bwrap --unshare-all
--ro-bind / / true` exits 0 (row 230's precheck line, F4); as `jarvis`, `/usr/bin/bwrap --version`
fails with "Permission denied" (P9); `getent group mosaic` lists no member and no other account has
gid `mosaic` (P9); `fs.protected_hardlinks` is 1 (P9). A role session running under the profile is F5.
One more test runs in 2a', because I can't settle it from here. Row 230's units keep the templates'
`NoNewPrivileges=yes`, and AppArmor limits profile changes for a task with no-new-privs set. The
`bwrap` attach happens on exec from an unconfined task, and the child's change is a stack
(`bwrap//&unpriv_bwrap`), which only narrows. I expect both to pass under no-new-privs, since bwrap
sets no-new-privs on itself before it runs the child, and Ubuntu's profile works with that. But I
haven't seen it run, and .53 is read-only to me. So as root, after step 4: `systemd-run --wait --pipe
--uid=mosaic --gid=mosaic -p NoNewPrivileges=yes -p ProtectSystem=strict -p ProtectHome=yes -p
PrivateTmp=yes /usr/bin/bwrap --unshare-user --unshare-pid --ro-bind / / cat /proc/self/attr/current`
prints `bwrap//&unpriv_bwrap (enforce)`. If it fails, the window stops at 2a' and rolls back 4 to 1,
before any `mosaic` step. That beats finding it at F5.
The drift stop (B2, Duckworth's REVIEW-r3 B2, mos-1098 point 1). r4 installed a profile file that no
package owns, from a pinned deb, onto a host where unattended-upgrades changes packages every day
(section 0). Three things can move under it: bubblewrap (its security updates install unattended), the
`apparmor` package and its parser, and any new profile a package ships that attaches to
`/usr/bin/bwrap`. Two more host values matter to this design: the user-namespace sysctl and the port
floor. So:
- `scripts/case1-pins.sh` holds the pins, sourced by the scripts below: the `apparmor-profiles` version
the file came from, the file's sha256, and the bubblewrap version 2a' installs. Changing a pin is a
reviewed commit and its own round. The bubblewrap version is an install pin, not a drift item:
a security update to bubblewrap keeps the statoverride, and the profile attaches by path.
- `scripts/case1-hostcheck.sh` checks seven items and prints one line each:
1. `/etc/apparmor.d/bwrap-userns-restrict` has the pinned sha256, and `dpkg -S` finds no package
that owns it.
2. `aa-status --json` lists `bwrap` and `unpriv_bwrap` in enforce mode, and no loaded profile named
`mosaic-bwrap`.
3. `grep -rlw bwrap /etc/apparmor.d` lists exactly `/etc/apparmor.d/bwrap-userns-restrict`. The word
match also catches a path written as `/{usr/,}bin/bwrap`. Section 0's read found no file today.
4. The live label: 2a' line `systemd-run --wait --pipe --uid=mosaic --gid=mosaic -p NoNewPrivileges=yes
-p ProtectSystem=strict -p ProtectHome=yes -p PrivateTmp=yes /usr/bin/bwrap --unshare-user
--unshare-pid --ro-bind / / cat /proc/self/attr/current` prints exactly `bwrap//&unpriv_bwrap
(enforce)`. It runs under the units' no-new-privs, so it reads the label a role session gets.
This catches a profile that attaches by a glob that item 3's search misses.
5. `dpkg-statoverride --list /usr/bin/bwrap` reads `root mosaic 0750`, and `stat` agrees.
6. `kernel.apparmor_restrict_unprivileged_userns` is 1 and `fs.protected_hardlinks` is 1.
7. `net.ipv4.ip_unprivileged_port_start` is above the Console's `--port` in the installed unit.
`--check` exits 0 when all hold and 1 otherwise. `--enforce` runs the same check and always exits 0.
On a failure, or while `/etc/mosaic-stack/DRIFT` exists, it applies the stop:
`apparmor_parser -R` on the profile if loaded, then `chown root:root` and `chmod 0700` on
`/usr/bin/bwrap`. It writes the failed items to `DRIFT` and appends a line to
`/var/log/mosaic-stack/hostcheck.log`. With bwrap closed to `mosaic`, every role session refuses
to start. That's the fail-closed state, and the stack needs no change for it. The chmod is undone by
the next bubblewrap upgrade, which reapplies the statoverride. But the apt hook runs again right
after that upgrade and finds `DRIFT`, so it closes bwrap again. Only root removes `DRIFT`, in a
reviewed step with a new pin or a Mos go. I don't call `dpkg-statoverride` from inside the hook,
because I haven't confirmed it can take dpkg's lock while apt holds the frontend lock.
- Three callers. A1 of every re-pin runs `--check` and refuses `profile-drift` on exit 1. The apt hook
is `DPkg::Post-Invoke { "/usr/local/lib/mosaic-stack/case1-hostcheck.sh --enforce || true"; };`, so it
runs after every package run, unattended ones included, and can't fail one. The boot unit,
`mosaic-hostcheck.service` (`Type=oneshot`, `After=apparmor.service`,
`Before=mosaic-bus@mosaic-stack.service`, `WantedBy=multi-user.target`), runs `--enforce` before the
bus starts, which covers a kernel or parser change that takes effect at boot.
- Owner. ops-01 owns the host side under a Mos go. It reads `hostcheck.log` in each STATE report and
sends any `DRIFT` to Mos. Before each re-pin window it also runs `--upstream`, which downloads the
current `apparmor-profiles` candidate into a temp dir and compares its `bwrap-userns-restrict` to the
pin. A difference goes to Mos as a new-pin request and refuses nothing. The pins and the scripts are
this row's, and later changes go through review like any other.
Who runs what:
| Step | Runs as | Through |
|---|---|---|
| 2a: `useradd --system --user-group --home-dir /srv/mosaic --shell /usr/sbin/nologin mosaic`, `/srv/mosaic` 0700, `/etc/mosaic-stack` `root:root` 0755, `/var/log/mosaic-stack` `root:root` 0750 | root | ops-01, `ssh localadmin@10.1.1.53`, `sudo` (#423 step 1) |
| 2a': node under `/opt/mosaic-stack`; bubblewrap, the AppArmor profile, the statoverride, the drift stop and the managed settings file, each with the test and rollback above; the no-new-privs test | root | ops-01, same path, R27 (mos-1091) |
| 2a'': a `git bundle` of the reviewed commit copied to .53; root clones it to `/root/mosaic-stack` and checks HEAD is the commit the go names; it installs the bundle as `/srv/mosaic/incoming/<sha>.bundle`, owner `mosaic`, 0600 | root | ops-01, same path |
| 2b: claude under `~/.local` (above); `git clone` of the bundle to `/srv/mosaic/src/mosaic-stack`; `scripts/bootstrap.sh`; `stack-copy.sh copy`; the install with the inputs above (C2-13b's `case2-install.sh --profile factory` if it has landed, else the run sheet's steps) | `mosaic` | `sudo -u mosaic -H env PATH=...` from `localadmin` (PATH above). `mosaic` gets no key and no login shell |
| 2b': place `claude-session.env` | Jason, or who he names (Q173 secrets rule) | section 3 |
| 2c: write `case1-unit.args`; `case2-unit.sh` with those arguments and `--copy <id>`, from `/root/mosaic-stack`; then `case2-precheck.sh --profile factory` as `mosaic`, which reads the Console's `--port` from the installed unit for its port-floor line, so it runs after the render; `systemctl enable` the bus unit; the first start of UNITS. The Console unit has no `[Install]` until Jason rules (C2-13a), so it isn't enabled, and after a host reboot root starts it by hand until then. The run sheet says so | root | ops-01, same path |
| 2c': `stack-release.sh adopt mosaic-stack --units external`, which records the running copy as the pointer (`do_adopt`, `scripts/stack-release.sh:856`) | `mosaic` | `sudo -u mosaic -H` |
| every later re-pin or rollback | root, with its `mosaic` steps through `sudo -u mosaic -H` | ops-01 runs `case1-repin.sh` (section 4) |
| acceptance (section 5) | `localadmin`, with `sudo -u mosaic -H` for the in-instance checks | the agent the window names |
No git credential lands on .53. Both checkouts come from bundles that ops-01 copies in, and a re-pin
brings a new bundle that `mosaic` fetches into its checkout.
## 2. Where Case 1 differs from Case 2
| Case 2 item (row 230 brief) | Case 1 | Reason |
|---|---|---|
| business file with the library mapping | business `mosaic-stack` with no `library` key; `--profile factory` refuses a `kind: "library"` project (`profile-library-business`) | the factory has no books repo |
| `bean`, `chet`, persona vars and files, specialists | factory roles from `roles/`, with a chief that has a harness var | Jason names the set in the go. I haven't read his live business file and won't. The acceptance needs only the chief (F3), since a chief with no harness var answers `no-route`. Others are added later by the human with `instances.set`, no new row |
| token slot `library-gitea` | none | the factory's per-agent git identity is row 199's credential store (Megavolt, #1697). Until it lands, git work stays on dragon-lin (Q182) |
| precheck: `bus.tester` is not on | `bus.tester` is on; `factory` requires it | the acceptance and the Q173 test system need the tester (agents act as tester, never Jason) |
| precheck: persona file modes and digests | skipped under `factory` | no persona files |
| precheck: `library.sync`, `library.status` | skipped under `factory` | no library |
| precheck: `ZAI_API_KEY` in the key file | doesn't run | no instance has provider `zai` |
| precheck: no Discord binding | kept | the factory's Discord connector stays on dragon-lin. Two connectors on one bot token would split its events |
| precheck: claude marker prompt with `--system-prompt-file` | runs, since the chief's harness is claude-code | |
| precheck: `sudo -n true` fails, `id -G` one entry, run root 0700, config validates, node 24 or later, `bwrap` exits 0, free memory, HOME | kept | node and bwrap are in row 230 (gosalyn-0034) |
| Console serve flags: `--port` 1..1023, `--business`, the preview trio, `--send`, the proof pair, no `--tester-port` | `--port 732`, `--business mosaic-stack`, `--tester-port 7333`. No preview trio, no `--send`, no proof pair | below |
| bus: `@PM_FLAGS@` and `@CLAUDE_ENV@` render empty | ` --pm`, and `Environment=MOSAIC_CLAUDE_BIN=/srv/mosaic/.local/bin/claude` | below |
All of these are inputs. Nothing in the Case 2 scripts hardcodes a Case 1 value (gosalyn-0034 item 4).
Row 230's brief says the factory profile takes the library serve set plus an optional `--tester-port`,
and that row 195 names any other difference (gosalyn-0049). .53 can't meet the library set, and the bus
needs two lines the brief renders empty. The factory profile I need from row 230:
1. Serve flags. `--port N` with N in 1..1023 and `--business ID` are required, and `--tester-port N` in
1024..65535, not equal to `--port`, is optional. The preview trio (`--public-host`, `--auth-header`,
`--allow-user`) and the proof pair (`--proof-header`, `--proof-file`) are refused. Both exist for a
proxy in front of the Console: the proxy signs the user in and adds the header, and the proof shows
the request came through it. .53 has no proxy. Jason reaches the Console with `ssh -L` (sage-1451
Q3), so there is no header to trust and no proxy to hold the proof. `--send` is refused too, since
serve takes it only with the trio or `--sso` (`packages/webui/src/cli.mjs:15`). When Q175 SSO reaches
.53, `--sso` with an optional `--send` (row 192) would be the factory's sign-in, and adding it is a
re-install with its own go. Row 230 can accept `--sso [--send]` under factory now or later. Either
way it needs a `console.sso` block in `mosaic`'s config, which this row doesn't write. `--checkout`
and `--board` stay refused, as in the library set.
2. `--pm` (factory only) renders the bus unit's `@PM_FLAGS@` as ` --pm`. The factory's PM dispatches its
work on .53 as on dragon-lin (section 4). With the token empty the bus never launches the PM, and
A5's PM check fails every re-pin.
3. `--claude PATH` (factory only) renders `@CLAUDE_ENV@` as `Environment=MOSAIC_CLAUDE_BIN=PATH`, the
line `scripts/bus-service.sh:79` renders today. PATH must be absolute. Without it the bus runs
`claude` from the unit's PATH, which is the node dir, `/usr/local/bin`, `/usr/bin` and `/bin`, and
holds no claude on .53. The chief's first launch would then fail. The precheck's claude marker
prompt should run the same binary, so it reads the path from the installed bus unit.
4. Row 190's two template tokens, if row 190 lands before row 230. Row 190 adds `@CONFIG_ENV@` to both
templates and `@SESSION_ENV@` to the bus template's `EnvironmentFile=` line. For the default
instance, `@CONFIG_ENV@` is deleted (`scripts/instance.sh`) and `@SESSION_ENV@` is
`<home>/.config/mosaic-dev/secrets/claude-session.env`. Row 230's "every token has one value" list
and its `EnvironmentFile=` check need both. This applies to the library profile too.
I've sent these to Gosalyn. If row 230's candidate lacks any of them, row 195's manifest names the
change to `case2-unit.sh` or `case2-precheck.sh`, reviewed by Duckworth with Webby told.
## 3. Model access
Q185: Case 1 uses the account the factory uses today, the Claude subscription behind dragon-lin's
`claude-session.env`. The install never copies that file from dragon-lin. In the window, Jason (or the
person he names under the Q173 secrets rule) makes a credential for .53 from that account and writes it
to `/srv/mosaic/.config/mosaic-dev/secrets/claude-session.env` as `mosaic`, mode 0600. Nobody prints it,
and no script reads its value. The precheck checks only its mode, owner and that the variable name is
present (a `grep -c` on the name, output a count).
Quota: the factory on dragon-lin and on .53 share one subscription. Q185's "use existing" accepts that.
Case 2 shares nothing with it (Mos's reading of Q185).
The pi harness stays off on .53 until P19 #383 settles (mos-0865). The business file names claude only.
## 4. Re-pin, rollback and backup on .53
Unit form (ruled, sage-1451 Q1): system units that root writes, with `User=mosaic`. `mosaic` can't
start or stop them, and root never runs code that `mosaic` can write. `stack-release.sh activate` today
does everything in one run under `systemctl --user` and puts the old stack back inside that run
(`rollback_in_run`, `scripts/stack-release.sh:561`). On .53 the same run needs two users. So one
re-pin is one ordered list, and one root command drives all of it: `scripts/case1-repin.sh`, run by
ops-01 from `/root/mosaic-stack`. Each `mosaic` step is `sudo -u mosaic -H` running
`/srv/mosaic/src/mosaic-stack/scripts/stack-release.sh` with `--units external`. Copies carry no
`stack-release.sh` (`COPY_PATHS`, `scripts/stack-copy.sh:23-30`), so the `mosaic` steps run from that
checkout, as dragon-lin's run from the canonical one.
What root holds, and where each value comes from (r4's B1):
| Value | Source |
|---|---|
| ID, the target copy | the go |
| TO, a rollback's target | the go, as `--to ID` |
| `--restore ACT --events N` | the go, when the rollback restores |
| OLD, the running copy | the `ExecStart=` of the two unit files root installed. If the bus and Console units name different copies, A1 refuses `units-disagree` |
| the `case2-unit.sh` arguments | `/etc/mosaic-stack/case1-unit.args` (section 1) |
| each `mosaic` step's result | its exit code |
Root parses nothing `mosaic` wrote and runs nothing under `/srv/mosaic` except through `sudo -u mosaic
-H`. It never needs the activation id: the `mosaic` steps find it themselves.
One result line per activation (B1). The stack log closes an activation on the first line after
`started` that names it, whatever the result (`interrupted`, :146-156). r4 wrote an interim
`... rollback pending` line at A3's and A5's failures. That line closed the activation, so R2 and R4
then refused `open-activation` against an activation the log no longer held open. The rollback could
never run as designed, and nothing tested the steps in sequence. r5 keeps an activation open from A3's
`started` line until its single result line, and A5 or R4 writes that line:
| Activation ends at | Result line | Written by |
|---|---|---|
| A3 fails before its rollback set exists (marker or capture) | `refused code marker-failed` or `refused code capture-failed`, as `capture || refuse` does today (:987). Nothing has changed, so there is nothing to roll back | A3 itself, exit 3, after it removes the marker. Root runs A2f |
| A5 passes | `ok` | A5 |
| A3 (after capture), A4 or A5 fails, and the rollback passes | `failed code <c> rollback ok` for A3 and A4, `gate-failed code <c> rollback ok` for A5 | R4 |
| the rollback fails at R4's gate | `rollback-failed code <c> rollbackCode <r>` | R4 |
| R1, R2 or R3 fails, or the run dies | none. The activation stays open, and the next activate refuses `interrupted-run` (:603-606), which Jason or Mos settles as on dragon-lin | |
What the result line carries (r5's B1). `log_line` (:58) writes `op "$OP"`, and `from_log` (:169)
reads only lines whose op is `adopt`, `activate` or `rollback`. So `confirm` doesn't log under its own
op. Before A5's `ok` line or any line R4 writes, it sets OP, FROM and TO from the open activation's
`started` line, ACT to that activation, and BACKUP from the rollback set's `BACKUP` file, which A3
writes when `backup_store` returns. A re-pin's lines then read `op activate`, a rollback's `op
rollback`, each with the from, to and backup a dragon-lin run would log. `from_log previous` and
`from_log backup` read .53's log as they read dragon-lin's. `confirm` reads only the line A3 wrote for
that activation, which `interrupted` already names.
A failing `mosaic` step after `started` writes no log line. It writes its code to `FAILED` beside
`AWAITING-UNITS` and exits 1. R4 reads that code for the result line. Root passes R4 `--cause
A3|A4|A5`, the step it saw fail, from its own exit codes. A4 is root's own step and has no `FAILED`
file, so R4 writes `failed code units-start` for it. A cause with no `FAILED` file where one should be
writes code `unrecorded`.
The open activation. A3 writes `$STATE/stack-rollback/<ACT>/AWAITING-UNITS` holding the target copy
(renamed from r3's `PENDING`, which read like the pointer's `pmPending`, `P_PENDING` at :128; N3).
`confirm` and `revert` take the copy root passed, not an activation id. Each lists the
`AWAITING-UNITS` files and the `interrupted` set (:146), and goes on only when there is exactly one
file, its activation is the one `interrupted` names, and the copy in it is the copy root passed.
Anything else refuses `open-activation` with exit 3 and the list it found, and changes nothing.
`confirm` without `--rollback` also refuses `activation-failed` when `FAILED` exists, so a second
confirm can't pass an activation that already failed. Whichever `confirm` writes the result line
removes `AWAITING-UNITS` and `FAILED` after it.
Refusals of the three external verbs. `confirm`, `revert` and `gate` act on an open activation by
design, so they skip the `interrupted-run` refusal (:605). Their refusals log `result refused` with no
activation id: they set ACT only after every check has passed. So a refused `confirm`, a wrong copy
say, leaves the open activation open. Today's `refuse` (:62-67) adds the activation id whenever ACT is
set, and the external verbs keep that order. A test and a mutant hold it (section 7).
Root's step log. Every step `case1-repin.sh` runs appends one line to `/var/log/mosaic-stack/case1-repin.log`,
`root:root` 0640: the time from `date -u`, the step, its argv, its exit code. `mosaic` can't write it,
so it stays readable after a run that dies before `mosaic` logs anything.
### Activate: `case1-repin.sh activate ID [--migrate] [--timeout S]`
| # | Step | Runner | On failure |
|---|---|---|---|
| A1 | `/root/mosaic-stack` HEAD is the commit the go names. ID is 40 lowercase hex and names a copy under `/srv/mosaic/.local/share/mosaic-stack/copies` that holds `scripts/mosaic` (row 230's copy-id rule). `case1-hostcheck.sh --check` exits 0, else `profile-drift` (B2). Root reads OLD from the unit files. `stack-release.sh activate ID mosaic-stack --units external --plan --expect-from OLD` exits 0. The plan refuses `pointer-differs` when the pointer's copy isn't OLD (N5), so a pointer and units that disagree stop before any stop | root; the plan as `mosaic` | refuse, exit 3. Nothing has changed |
| A2 | `systemctl stop` UNITS in reverse order, then wait until each is inactive, within `--timeout` | root | A2f below |
| A3 | `stack-release.sh activate ID mosaic-stack --units external [--migrate]`. Before it logs `started` (N3), it refuses `units-running` unless each of UNITS is inactive (system-manager reads, which need no privilege) and the store is stopped (`stopped`, :359), and `interrupted-run` while any activation is open. Then, in this order: logs `result started`; makes `$RS` with capture's `mkdir -p` of `stack-rollback` and exclusive `mkdir -m 0700` (:415-416), and writes `AWAITING-UNITS` in it at once (N1); captures the rollback set into that directory, making none (pointer, wrapper, and the sha256 and `ExecStart=` of each installed unit, read only); takes `backup_store` (:372) into `$DATA/backups/stack/<ACT>` and writes its name to `BACKUP` in the rollback set; under `--migrate`, writes `MIGRATING` in the rollback set and sets TAKEN, both before `migrate-schema.mjs` runs, where today's run sets TAKEN (:1004-1005, r5's B2), then migrates; installs the wrapper for ID. It writes no pointer | `mosaic` | a refusal before `started` exits 3 with nothing changed. A marker or capture failure, an existing `$RS` included, removes the marker if it wrote one, logs its `refused` line and exits 3. Root runs A2f for both. Any later failure writes `FAILED` with its code, logs nothing, cleans up nothing and exits 1. Root runs R1-R4 with `--cause A3` |
| A4 | `case2-unit.sh` with `case1-unit.args` and `--copy ID`, `systemctl daemon-reload`, `systemctl start` UNITS in order | root | R1-R4 with `--cause A4` |
| A5 | `stack-release.sh confirm ID mosaic-stack --units external --timeout S`. This is `gate` (:522), which polls `gate_once` (:471) every second until `--timeout`, with the external checks below. On a pass it writes the pointer, logs `result ok`, removes `AWAITING-UNITS`, exit 0. When the timeout runs out it writes `FAILED` with the last gate code, logs nothing and exits 1 | `mosaic` | R1-R4 with `--cause A5` |
Why A3 writes the marker first and does no cleanup of its own. With the marker written last, as in r4,
an A3 failure after `started` left an open activation with no marker, and R2 and R4 refused it. With
the marker first, every A3 failure after the marker leaves exactly what R2 needs: the open activation,
its marker and whatever part of the rollback set exists. R2 is then the one cleanup path, for A3, A4
and A5 alike. Two cleanup paths, A3's own and R2, would each need their own tests for each failure point.
A2f, a stop that fails or times out, or an A3 that exits 3. This run has no open activation: A2 logs nothing, and
A3's exit 3 comes either before `started` or after its own `refused` line. Nothing of `mosaic`'s has
changed. Root reads `date -u +%s` as T, runs `systemctl start` UNITS in order, then `stack-release.sh gate
OLD mosaic-stack --units external --since T --timeout S`, a new verb that runs `gate` on OLD and writes no
log line, no pointer and no file. A2f has no `started` line to take a start floor from (the first
re-pin after adopt has none at all), so `--since` is the floor, and `gate` refuses `usage` without it
(r5's N2). T is root's own clock, so root still parses nothing `mosaic` wrote. A unit that never
stopped keeps its old `ActiveEnterTimestamp`, fails `stale-start` and takes the exit 2 path below.
If that passes, root logs `A2 failed, OLD running` (or `A3 refused`) in its step log and exits
1. If it fails, root stops UNITS, logs `OLD gate failed <code>` and exits 2. The next run's
`interrupted` check finds nothing open either way.
The external gate. `gate_once` keeps its order and codes, with these changes under `--units external`:
- `systemctl` without `--user` for `is-active` and `show`. Both work unprivileged against the system
manager.
- This start's time is the bus unit's `ActiveEnterTimestamp`, read with `systemctl show --timestamp=unix
-p ActiveEnterTimestamp`, in place of the `START_MS` that `start_both` (:451) sets on dragon-lin.
It is whole seconds (`@1792540800`), and the `started` line has milliseconds (N4). So the rule
compares seconds: the start passes when its second is at or after the floor of `started`'s second.
A unit that was active across A3 can't reach this check, because A3 refuses `units-running`, so the
equal second can only be a start after A3. An earlier second fails `gate_once` with the gate code
`stale-start`. It's a gate failure like any other, not a refusal (r4 had it as a refusal, which
would have logged against the open activation): `gate` keeps polling, since `Restart=on-failure`
can bring a fresh start, and at the timeout `FAILED` holds `stale-start`. The PM check's
`startedAt` bound uses the same second.
- `copy-check` runs `stack-copy.sh check --units external`, which reads the system units.
- The PM check is below.
The PM (r4's B2). .53 runs the PM: the factory's PM dispatches its work, as it does on dragon-lin, and the
serve flags carry `--pm`. The pointer's `pm` comes from the installed bus unit's `ExecStart=`, as the
unit parse at :195 already reads it. Today's `pm_launched` (:505) takes the run id from the
`launched ...` line in `journalctl --user -u BUS_UNIT`. On .53 that read fails twice over: the unit is
a system unit, and under `SplitMode=uid` the logs of a system-range uid land in the system journal,
which `mosaic` (no `adm` or `systemd-journal` group) can't read. Under `--units external` the check
reads `launches list --json` alone. It passes when exactly one entry has business `mosaic-stack`,
instance equal to the business file's `launch.by` (the PM's instance, the one `bus start --pm` launches
at `packages/cli/src/launcher.mjs:675`), `live` true, and `startedAt` at or after this start's time.
Zero entries, two, a stale one or one for another business fail with `pm-launch-unconfirmed` and what
it saw, as today. A run the PM starts for another role has another instance, so it doesn't count.
Why not the other reads. A regular uid for `mosaic` would put its logs in a per-user journal it can
read, but it changes the account Case 2's install creates. Root reading the journal and passing the run id in
would make root parse output from `mosaic`'s process, which section 4's rule forbids.
Under `--units external`, `pm-on` (:1137) and `activate --pm-after-harness` refuse `external-pm`. Both
flip the PM on dragon-lin by restarting the user bus with new arguments. On .53 the PM's state is a
serve flag in a root-owned unit, so changing it is a re-install with its own go.
### Rollback inside the run (R1-R4)
| # | Step | Runner | On failure |
|---|---|---|---|
| R1 | `systemctl stop` UNITS, then wait as in A2 | root | rollback-failed |
| R2 | `stack-release.sh revert ID mosaic-stack --units external`. It finds the open activation as above, puts the wrapper back from the rollback set and checks its sha256, and restores TAKEN (`restore_store`, :396) only if the rollback set holds `MIGRATING`. Each part is idempotent, so it works for an A3 that stopped partway: no `MIGRATING` means no restore, a migrate that failed partway is restored from the backup taken before it, and copying the saved wrapper over an unchanged one changes nothing. It writes no log line | `mosaic` | rollback-failed, activation left open |
| R3 | `case2-unit.sh` with `case1-unit.args` and `--copy OLD`, `systemctl daemon-reload`, `systemctl start` UNITS | root | rollback-failed, activation left open |
| R4 | `stack-release.sh confirm ID mosaic-stack --units external --rollback OLD --cause A3\|A4\|A5 --timeout S`. It finds the open activation, checks OLD is its `from`, gates OLD with `gate`, and writes the activation's one result line from the cause and the `FAILED` code: `failed code <c> rollback ok` (A3, A4), `gate-failed code <c> rollback ok` (A5), or `rollback-failed code <c> rollbackCode <gate>` when OLD's gate fails | `mosaic` | rollback-failed, logged |
rollback-failed: root stops UNITS and exits 2, as `rollback_failed` (:587) does. When R4's gate fails,
the log has the `rollback-failed` line and the activation is closed. When R1, R2 or R3 fails, no
`mosaic` step writes a result, so the activation stays open with its marker. Root's step log has the
codes, the next activate refuses `interrupted-run` until Jason or Mos settles it (:605), and status
shows it. I chose that over a log-only verb for root to call after an R2 failure: nothing could gate
the old stack at that point, and a closed activation would let the next activate start against a
wrapper and store in an unknown state. Exit codes match `stack-release.sh`: 0 ok, 1 failed with the old
stack running (rolled back, or A2f), 2 rollback-failed with UNITS stopped, 3 refused with nothing
changed.
### Rollback to the previous copy: `case1-repin.sh rollback --to ID [--restore ACT --events N]`
The same list with TO in place of ID. The go names TO, from Mos's or Sage's read of `stack-release.sh
status` as `mosaic`. Root passes it. A1's plan, `stack-release.sh rollback mosaic-stack --units external
--to ID --plan --expect-from OLD`, refuses `pointer-differs` as in activate. It also exits 3 unless ID
is `from_log previous` (:162), the previous distinct copy, so a
stale or mistyped go refuses before anything stops. With `--restore`, the plan also refuses unless
`$DATA/backups/stack/ACT` exists. `--to` is a new flag, accepted only with `--units external`;
dragon-lin's rollback keeps reading TO from the log.
The post-backup event count runs in A1's plan as `mosaic` while the bus still holds the store (the
`heldcount` read, :1074), and again in A3 on the stopped store. A difference refuses
`post-backup-events`, as `run_steps` does (:1013).
### Migrations (N4)
`--migrate` under `--units external` runs only inside A3, and A3 refuses `units-running` before it logs
or migrates anything while any of UNITS is active. So a migration never runs against a store a live bus
holds. The test is in section 7.
### Backup
A3's `backup_store` is today's. It copies the stopped store into `$DATA/backups/stack/<ACT>`, so
`rollback --restore <ACT>` and row 181's guard find it. As on dragon-lin, the config dir isn't in the
backup, and the credential isn't either. PBS backups of VM 1008 cover the disk (P4). A restore swaps
only `bus/` and leaves `secrets/claude-session.env` in place (N6, in the run sheet).
The install's adopt (2c') takes no backup. The store holds only what the acceptance wrote, and the first
re-pin's A3 takes the first backup.
### The log
Every `mosaic` line goes to `$DATA/state/stack-activation-log.jsonl`, as today, with `units external`
added. So `from_log previous` reads .53's history the same way it reads dragon-lin's. A3 writes
`started`. A5 or R4 writes the one result line, with the op, from, to and backup of that activation,
or A3 its `refused` line before any change. Refusals of
`confirm`, `revert` and `gate` log no activation id. If a run dies between `started` and the result,
the log keeps `started` with no result, the next run's `interrupted` check reports it, and root's step
log shows the last step that ran.
### Who
Mos or Sage plans each re-pin. ops-01 runs `case1-repin.sh` on Jason's or Mos's go, and a STATE report
follows. The gate and the in-run rollback have one owner, the root command, as they do on dragon-lin.
### dragon-lin's path
`--units external` is a new value. Without it, `stack-release.sh` and `stack-copy.sh check` run as they
do today, `pm_launched` still reads the user journal, and rollback still takes TO from the log.
`confirm`, `revert`, `gate` and `--to` refuse without it. Section 7 names the tests that show this.
### Not taken
- Row 190's user units under `mosaic`, with root running `loginctl enable-linger mosaic` once. Every
existing re-pin tool would work unchanged and root would be needed only once. But the unit form
would differ from Case 2, and a user manager would run as `mosaic`.
- A sudoers rule that lets `mosaic` start and stop the units. P1 and row 230's precheck require that
`mosaic` has no sudo.
- `confirm` and `revert` taking ACT from root. Root would have to read it from `mosaic`'s output or
log, which the rule above forbids (r4's B1).
## 5. Acceptance checks
A new script, `scripts/case1-accept.sh`, in this row's manifest. It runs on .53 as `localadmin` and prints
one line per check: `<id> <path> <mark> <evidence>`. The marks are `closed`, `accepted Q182`,
`accepted Q183`, `accepted mos-1096`, and `unnamed`, which means "not named in Q182, Q183 or a Mos
ruling". The script ends with a count of each mark. It exits non-zero only when a functional check (F)
fails. A P check never changes the exit status. Each `unnamed` line goes to Mos, who takes it to Jason
(R8 N1).
Functional checks. Any failure fails the row:
| Id | Check | How |
|---|---|---|
| F1 | the units run as `mosaic` | `systemctl show <unit> -p User -p MainPID` for each of UNITS; each MainPID's uid in `/proc/<pid>/status` is `mosaic`'s, and each unit is active |
| F2 | the Console listens on 127.0.0.1:732 on .53 and answers there | `curl -fsS http://127.0.0.1:732/healthz` on .53 succeeds, and the listener table shows 732 bound to 127.0.0.1 only, by the Console unit's MainPID. Jason reaches it with `ssh -L 7332:127.0.0.1:732 localadmin@10.1.1.53` and his browser at 127.0.0.1:7332 as today. A LAN listener waits for Q175 SSO (sage-1451 Q3) |
| F3 | one tester message gets a reply | as `mosaic`: `mosaic talk chief <marker> --tester`, then `mosaic follow --tester`, with a 300 s limit. A stored reply to the marked message passes. The inbox alone shows decisions only, so it isn't used (N3) |
| F4 | the precheck passes | `case2-precheck.sh --profile factory` exits 0 as `mosaic` |
| F5 | a role session runs under bwrap with the profile | while F3's reply is pending: a `/usr/bin/bwrap` process owned by `mosaic` has a claude process under it. `/proc/<bwrap pid>/attr/current` equals `bwrap (enforce)` and the claude process's equals `bwrap//&unpriv_bwrap (enforce)`, each compared as an exact string after the trailing newline is cut (N1). If `launches list --json` records the sandbox, the build reads that for the first part, since a short session can end before `ps` sees it |
| F6 | claude is the pinned version | `/srv/mosaic/.local/bin/claude --version` as `mosaic`, after F3, equals the run sheet's pin |
| F7 | the PM run is live | `launches list --json` as `mosaic` has exactly one live entry for the PM's instance, the check `confirm` runs (section 4) |
Privilege and host paths. These report and never fail the row:
| Id | Path | Mark | Evidence the script prints |
|---|---|---|---|
| P1 | `mosaic` has sudo | closed only when `sudo -l -U mosaic` prints "is not allowed to run sudo" (N2); any other output is unnamed | the `sudo -l` result |
| P2 | `mosaic` has supplementary groups or a key | closed when `id -G mosaic` is one entry and `~mosaic/.ssh/authorized_keys` is absent or empty | `id`, file state |
| P3 | `localadmin` NOPASSWD sudo, reached with the dragon-lin key | accepted Q183 (agents keep ssh to .53; localadmin from #423 step 1) | `sudo -l -U localadmin` |
| P4 | root on pve over VM 1008, and PBS backups of VM 1008, which hold `/srv/mosaic` including the credential | accepted Q182 and Q183 (agents keep root on pve and PBS) | a fixed line; the script can't see pve from inside |
| P5 | `jarvis`, which runs Hermes, can reach `/srv/mosaic` through sudo, or through the docker or lxd groups, which are root-equivalent | closed when `jarvis` is in none of sudo, docker and lxd; unnamed otherwise. #423 step 4 lands before the install (mos-1091 4c), and its `step4.sh` removes `jarvis` from sudo, docker, lxd and adm, so after it this reads closed (Duckworth N2) | `id jarvis` |
| P6 | other members of sudo, docker, lxd | closed when sudo holds only `localadmin` and docker and lxd are empty. Every other member is its own unnamed line, `jarvis` included (N1) | `getent group` lines |
| P7 | a file under `/etc/sudoers.d` naming `mosaic` | closed when none | the file names |
| P8 | the Console or tester reachable from the LAN | closed when 732 and 7333 are bound to 127.0.0.1 only, each by the Console's MainPID, and nothing listens on 7331 | listener table |
| P9 | a user namespace through `/usr/bin/bwrap` | accepted mos-1096 only while all seven hold, unnamed otherwise: (1) `dpkg-statoverride --list /usr/bin/bwrap` reads `root mosaic 0750`; (2) `aa-status --json` lists `bwrap` and `unpriv_bwrap` in enforce mode; (3) `/etc/apparmor.d/bwrap-userns-restrict` has the pinned sha256 and no package owns it; (4) `sudo -u jarvis /usr/bin/bwrap --version` fails with "Permission denied"; (5) `getent group mosaic` lists no member and no other account in `/etc/passwd` has its gid; (6) `fs.protected_hardlinks` is 1; (7) no `mosaic-bwrap` profile is loaded, no other file under `/etc/apparmor.d` names `bwrap`, and the live label of a process bwrap runs is exactly `bwrap//&unpriv_bwrap (enforce)` (N1, mos-1098 point 2). Items 1-3, 6 and 7 are `case1-hostcheck.sh --check`'s, so P9 reuses its lines | each item's line, and Mos's reason: "Case 1's agents are the factory's agents. In Q182 and Q183 Jason accepted that the factory reaches root on pve and, through it, every VM. So a user-namespace path to root on .53 gives that class no new reach." With the profile, what's left is bwrap's own setup, run by `mosaic` |
| P10 | a session binds the tester port, 7333, while the Console is down | accepted mos-1101 while `ip_unprivileged_port_start` is above 732 and 7333 is held by the Console's MainPID at the check. Serve refuses a tester port below 1024 (`cli.mjs:33`), so the Console-port fix can't cover it. A squatter would get the tester's writes and could answer them. The tester's users are the factory's agents, the class Mos's reading covers (mos-1101 item 5). Mos can object, and then this line reads unnamed | the floor, and the owner pid of 7333 |
Tests, in `packages/cli/tests/case1-accept.test.mjs`. The script takes a fake root (`--root DIR`) for
`/proc`, `/etc` and `getent` output, and fake `systemctl`, `curl`, `ss`, `sudo`, `bwrap`, `aa-status`,
`dpkg-statoverride`, `sysctl` and `claude` on PATH. Cases:
- each F check failing gives a non-zero exit with its line;
- each P check in each of its marks, including P1 with output that grants nothing but lacks the refusal
text (unnamed), P9 unnamed once for each of its seven items missing, and P10 unnamed with a floor of
700 and with 7333 held by another pid;
- F5 fails on a label of `unpriv_bwrap (enforce)`, `bwrap//&unpriv_bwrap (complain)` and
`bwrap (enforce)` with a trailing space;
- a run with every P check `unnamed` still exits 0;
- the summary counts match the lines.
Mutants: one per F check (the check made to pass on failure), one that maps P5 to `closed`, one that
closes P1 on any output without a rule line, one that drops P9's jarvis item, one that drops P9's item
7, and one that makes F5 a substring match.
Nothing in the tests reaches .53 or dragon-lin's live state.
## 6. Rulings
sage-1451:
- Q1, unit form: the brief's system unit. C2-13's `ExecStart=` check accepts a varying copy id, and
C2-13 took that (gosalyn-0034 item 3).
- Q3, the Console: it listens on 127.0.0.1 on .53 (`packages/webui/src/serve.mjs:466`; `cli.mjs` passes no
host), and Jason reaches it with `ssh -L`. F2 says so. A LAN listener waits for Q175 SSO.
- Q4, the date: if C2-13's candidate is late, I build on C2-13's latest commit and rebase (Q177). Under
the #426(a) push rules (mos-1086) a rebase goes to a new branch name.
mos-1091 (none goes to Jason):
- Q2 4a: installing bubblewrap is an install step, run by ops-01 as root under R27. Rollback is
`apt-get remove`.
- Q2 4b: a per-binary AppArmor profile for `/usr/bin/bwrap`, the sysctl left at 1. r4 takes Ubuntu's
`bwrap-userns-restrict` for it (mos-1096). Section 1 has the steps.
- 4c: P5 ends at #423 step 4, which Mos sent to Duckworth so it lands in Jason's VM 1008 boot session,
before the install. The P5 check stays.
mos-1096:
- B1 and B2 are mine, and so is the PM answer: .53 runs the PM (section 4).
- The `dpkg-statoverride` on `/usr/bin/bwrap`, the hardlink and group checks, the jarvis refusal and F5
under the profile (sections 1 and 5).
- P9 is Mos's acceptance, with the reason quoted in section 5.
- Jason's Q9 on install and upgrade is open. r4 goes ahead, since B1's re-pin and rollback are needed
whatever he answers. If his answer changes how upgrades reach .53, the change lands in section 4 and
the run sheet as a new round.
mos-1098 and mos-1101 (on Duckworth's REVIEW-r3):
- Duckworth's CHANGES stand. B1 and B2 are fixed in this round, with B1's real-sequence test, and all
three of B2's asks: an owner across upgrades, what fails closed, and the pins as reviewed values.
- r4's PM, statoverride and P9 stand. mos-1098's two questions are answered in section 1: the profile's
owner across upgrades is ops-01 with `case1-hostcheck.sh`, and the stop is bwrap closed to `mosaic`.
Both profiles attach to `/usr/bin/bwrap`: `bwrap` to the binary and `unpriv_bwrap` stacked on what it
runs. P9 item 7 checks that no third one does.
- P9 has seven items and F5 compares exact strings (N1).
- The port squat applies (section 1). The Console takes Case 2's fix, a port below 1024 with
`CAP_NET_BIND_SERVICE` and the floor checked. Only the tester port can't take it, and P10 records it
under Mos's reading that Case 1's sessions are the factory class. I've told Mos, who can object.
- Work on r5 goes ahead while Jason's Q9 (Vikunja #429) is open. If install and upgrade become
product commands, these steps become that installer's configuration and row 195 rebases on its row
(sage-1481, sage-1491).
sage-1499: C2-13a is row 230 (#1732, Gosalyn), and row 195 chains on row 230's candidate.
gosalyn-0049: the factory profile is the library set plus an optional `--tester-port`, and row 195 names
any other difference. Section 2 names four, with the reason .53 can't meet the preview trio, `--send` or
the proof pair. The managed-settings route needs no Case 2 change.
## 7. Manifest (planned)
- `scripts/case1-repin.sh` (new) and `packages/cli/tests/case1-repin.test.mjs` (new). Fake `systemctl`,
`sudo` and `case2-unit.sh` on PATH record their argv, with a scratch home standing in for `/srv/mosaic`,
a scratch dir for `/etc/mosaic-stack` and one for `/var/log/mosaic-stack`. Cases:
- a pass runs A1-A5 in order;
- a failure at A3, A4 or A5 runs R1-R4 with `--cause` set to that step, and exits 1;
- an A3 exit 3 runs A2f, not R1-R4;
- a failure at R2 or R4 exits 2 with UNITS stopped;
- A1 refuses `profile-drift` when a fake `case1-hostcheck.sh --check` exits 1, before any stop;
- every `mosaic` step's argv carries `env PATH=<node dir>:/usr/local/bin:/usr/bin:/bin`, with the node
dir from `case1-unit.args`;
- a HEAD mismatch, an ID that fails the copy-id rule, or unit files that name two copies refuses with
exit 3 before any stop;
- A1's plan argv carries `--expect-from` with OLD from the unit files;
- A2f passes `gate OLD --since T`, with T from a fake `date` that root calls before its start;
- A2f: a fake `systemctl` that never reports inactive. The script starts UNITS, runs `gate OLD`, exits
1 with OLD's gate passing and 2 with it failing, and the stack log is byte-identical before and
after;
- `rollback --to` with an ID that isn't `from_log previous` exits 3 before any stop;
- every exec of a path under the home goes through `sudo -u mosaic -H` (from the recorded argv);
- no argv root passes holds an activation id, and OLD comes from the unit files, not from anything
the `mosaic` steps print;
- every `case2-unit.sh` call carries the arguments in `case1-unit.args` and differs only in `--copy`;
- the step log has one line per step with its exit code.
- `scripts/stack-release.sh`: `--units external` for `adopt`, `activate`, `rollback` and `status`, the
`--to` flag for an external rollback, and the new `confirm`, `revert` and `gate` verbs, which refuse
without it. `scripts/stack-copy.sh`: `check --units external`, which reads the system units. Tests in
`packages/cli/tests/stack-release-external.test.mjs` (new), against a fake system-scope `systemctl`
and a fake `mosaic` for `bus status` and `launches list`:
- `units-running` refuses while any unit is active, and `activate --migrate` with a unit active
leaves the store, the log and TAKEN as they were (N4);
- activate leaves the pointer as it was, writes `AWAITING-UNITS`, and writes a backup of the stopped
store under `$DATA/backups/stack/<ACT>`;
- `confirm` and `revert` refuse `open-activation` with zero `AWAITING-UNITS` files, with two, with
one whose copy isn't the one passed, and with one whose activation isn't in the `interrupted` set;
- `confirm` polls: a fake bus whose status turns good after 3 s passes with `--timeout 10` and fails
with `--timeout 1`;
- `stale-start` (N4): with `started` at 1792540800.900, an `ActiveEnterTimestamp` of `@1792540800`
passes and `@1792540799` fails `gate_once` with `stale-start`. A fake that turns to `@1792540805`
after 2 s passes with `--timeout 10`, which shows it's a gate code that polling can clear, not a
refusal;
- `confirm` on a pass writes the pointer, logs `ok` and removes `AWAITING-UNITS`. On a timeout it
writes `FAILED` with the gate code, leaves the pointer, the log and the marker as they were, and
exits 1;
- `confirm` without `--rollback` refuses `activation-failed` when `FAILED` exists;
- every line `confirm` writes has op `activate` or `rollback`, never `confirm`, and the from, to and
activation of its `started` line, and backup equal to the rollback set's `BACKUP`;
- `gate` (N2): with `--since 1792540800`, `@1792540800` passes and `@1792540799` fails `stale-start`;
without `--since` it refuses `usage`;
- `--plan --expect-from` (N5) refuses `pointer-differs` when the pointer names another copy, for
activate and for `rollback --to`;
- A3's order: a fake that fails the capture leaves the log with `started` and `refused code
capture-failed` for that activation, no marker and no `FAILED`, and exit 3. An existing `$RS`
gives `refused code marker-failed` and exit 3, and a capture under `--units external` makes no
directory of its own. A fake that fails the
wrapper install after a migration leaves `started` only, the marker, `FAILED`, TAKEN as migrated,
and exit 1;
- refusals by `confirm`, `revert` and `gate` log with no activation id, and the `interrupted` set
is the same before and after each one;
- the PM check passes with one live entry for the PM's instance since the start, and fails
`pm-launch-unconfirmed` with none, with two, with one whose `startedAt` is before the start, with one
not live, and with one for another business. A second live entry for another instance doesn't
count. A fake `journalctl` on PATH records any call, and no external run makes one;
- `pm-on` and `activate --pm-after-harness` refuse `external-pm`;
- `revert` puts the wrapper back, and restores TAKEN only after a migration;
- `confirm --rollback OLD` writes the line for each `--cause` (A3 and A4 `failed`, A5 `gate-failed`,
with the `FAILED` code, `units-start` for A4, `unrecorded` when the file is missing), or
`rollback-failed` when OLD's gate fails. It removes the marker and `FAILED` after the line, and
refuses when OLD isn't the open activation's `from`;
- real sequences on one log (B1). The fake `sudo` execs the real `stack-release.sh` as the test's
user, so each step reads the log and state the last one wrote. Fakes stand in only for
`systemctl`, `mosaic` and `case2-unit.sh`. Each sequence ends with the log holding exactly one
`started` and one result line for its activation, an empty `interrupted` set, no marker or
`FAILED`, and a following `activate --plan` that doesn't refuse `interrupted-run`:
1. activate, confirm times out, revert, `confirm --rollback --cause A5` passes:
`gate-failed code <c> rollback ok`;
2. activate with `--migrate`, the wrapper install fails, revert restores TAKEN, `confirm --rollback
--cause A3`: `failed code <c> rollback ok`;
3. activate, root's A4 fails, revert, `confirm --rollback --cause A4`: `failed code units-start
rollback ok`;
4. activate whose capture fails: `refused code capture-failed` and nothing open, then A2f's
`gate OLD` writes nothing;
5. activate, then `confirm` with the wrong copy is refused, then revert and `confirm --rollback
--cause A5` still close the activation, because the refusal didn't;
6. activate, revert fails (a wrapper whose saved copy is missing): the activation stays open, and
the next `activate --plan` refuses `interrupted-run`;
7. adopt C0, activate C1 and confirm, activate C2 and confirm (r5's B1): both `ok` lines read
`op activate` with their `started` line's from and to and the backup each took. `rollback --to C1
--plan` exits 0 and `--to C0 --plan` refuses. A rollback to C1, confirmed, logs `op rollback`, and
`from_log previous` then names C2;
8. activate with `--migrate`, where the copy's `migrate-schema.mjs` is a fake that writes to the
store and exits 1 (r5's B2): `FAILED` holds `migrate-failed` and `MIGRATING` exists. revert
restores the store byte-identical to the backup, and `confirm --rollback --cause A3` gates OLD on
it: `failed code migrate-failed rollback ok`;
- `rollback --to` refuses when ID isn't `from_log previous`, and `rollback --restore` finds a backup
that an external activate took;
- `gate` writes nothing: the log, the pointer and the state dir are byte-identical before and after;
- `confirm`, `revert`, `gate` and `--to` refuse without `--units external`.
- dragon-lin unchanged. `stack-release.test.mjs` and `stack-copy.test.mjs` run unchanged and green. A
new case in `stack-release-external.test.mjs` runs a default activate, rollback and check against a
fake `systemctl` that records argv, and every call carries `--user`. A default `activate --plan`
prints the same bytes on the candidate as on its base, with the base output as a fixture.
- Mutants for the re-pin:
- `confirm` writes the pointer on a gate failure;
- `confirm` calls `gate_once` once in place of `gate`;
- the external PM check counts entries of any instance;
- the external PM check drops the `startedAt` bound;
- `confirm` takes the newest `AWAITING-UNITS` instead of refusing two;
- `rollback --to` skips the `from_log previous` comparison;
- `case1-repin.sh` skips R1-R4;
- `case1-repin.sh` exits 0 after A2f;
- `case1-repin.sh` runs `mosaic`'s `stack-release.sh` without `sudo -u mosaic`;
- the default path drops `--user` from one `systemctl` call;
- `confirm`'s timeout writes r4's interim `gate-failed ... rollback pending` line (B1; killed by
sequences 1 and 5);
- A3 writes `AWAITING-UNITS` last, after the wrapper (killed by sequence 2);
- an external verb's refusal logs with the open activation id (ACT set before the checks; killed by
sequence 5 and the refusal test);
- `stale-start` is a refusal again (killed by the polling case);
- `revert` restores TAKEN without `MIGRATING` (killed by the revert case);
- `confirm` logs with its own op, `confirm` (killed by sequence 7 and the op case);
- A3 writes `MIGRATING` after `migrate-schema.mjs` (killed by sequence 8);
- `gate` takes its floor from the newest `started` line when `--since` is missing (killed by the
`gate` case).
- `scripts/case1-pins.sh` and `scripts/case1-hostcheck.sh` (new),
`packages/cli/systemd/mosaic-hostcheck.service` (new) and `packages/cli/tests/case1-hostcheck.test.mjs`
(new). The script takes `--root DIR` for `/etc`, `/proc/sys` and `/usr/bin/bwrap`'s stat, and fake
`aa-status`, `dpkg`, `dpkg-statoverride`, `apparmor_parser`, `systemd-run`, `chown` and `chmod`
on PATH record their argv. Cases:
- `--check` exits 0 with all seven items and 1 with each one failing alone, its line naming it;
item 3 fails on a second file naming `bwrap` as a word, one that writes `/{usr/,}bin/bwrap` included,
and passes with `bwrap-userns-restrict` alone. Item 4 fails on a label of `unpriv_bwrap (enforce)`
or one with a trailing space, and its `systemd-run` argv carries `-p NoNewPrivileges=yes` and the
other 2a' properties. Item 7 fails with the floor at 732;
- `--enforce` exits 0 on a pass and on every failure. On a failure it unloads the profile, closes
bwrap to `root:root` 0700, writes `DRIFT` and one log line. With `DRIFT` present and every item
passing, it still closes bwrap;
- `--enforce` never calls `dpkg-statoverride` with `--update`, `--add` or `--remove`;
- a bubblewrap version other than the install pin is a printed line, not a failure;
- `--upstream` against a fake deb whose file differs prints the difference and exits 0;
- the service unit has `Type=oneshot`, `After=apparmor.service`,
`Before=mosaic-bus@mosaic-stack.service` and runs `--enforce`.
Mutants: `--enforce` exits 1 on a drift; `--enforce` ignores `DRIFT` when the items pass; item 1
drops the `dpkg -S` owner check; item 3 matches only the `/usr/bin/bwrap` path; item 4 matches a
substring; item 4 drops `-p NoNewPrivileges=yes`; item 7 compares with `>=`.
- `scripts/case1-accept.sh` (new) and `packages/cli/tests/case1-accept.test.mjs` (new), section 5.
- Row 230 (C2-13a). `--profile`, the node, bwrap and port-floor precheck lines, the copy-id rule, the
two units and the Console's capability lines are in its brief at 5b8def2f1. Section 2's four items
are not. Any of them that row 230's candidate lacks becomes a named change here, reviewed by
Duckworth with Webby told. I told Gosalyn earlier that the settings-file route for
`DISABLE_AUTOUPDATER` doesn't reach a role session, and that Case 1 uses the managed settings file.
- Docs: the Case 1 run sheet `docs/guides/case1-install.md` (new). It lists the window's steps in order,
each with who runs it and its go, the pinned node sha256 and claude version, a pointer to
`case1-pins.sh` for the bubblewrap and profile pins, the statoverride, the drift stop and the managed
settings file, each with its rollback, how ops-01 clears a `DRIFT`, the `ssh -L` line for port 732,
and a note that a restore leaves the credential in place.
- No change to `~/.mosaic-dev`, `~/.config/mosaic-dev`, dragon-lin's units or `~/.mosaic`.
+9
View File
@@ -0,0 +1,9 @@
# Gizmoduck
Design and builder seat, Claude Opus 5.5, a T3 thread (9312a896). Author of the federation design
(`agents/gizmoduck/work/federation/DESIGN.md` on `gizmoduck/federation-design`, #414). Builds FED R1, the
access.json loader and validator, under Mos's order mos-1023 item 1.
No launcher in this directory. The seat works in its own worktree and has no credentials of its own; until
the credentials row settles per-agent identity, it pushes only its own `gizmoduck/*` branches through the
repository credential helper (mos-0981), never `refactor`, `main` or another seat's branch.
@@ -0,0 +1,158 @@
# PRD 1.7 draft: the chief of staff's hire and fire authority, the coder's push to its fork, and agent
# host access through the credential store
Draft by Honker for row 194 (#1692), under Mos's ruling mos-1017 (through Sage, sage-1365). Not approved.
It changes nothing until Jason approves it as a `prd.approve` decision in the Console. Sage opens that
decision and sends Mos its id. Base: `docs/prd/mosaic-stack.md` version 1.6 at 53bb59462.
## What it changes
Row 194 adds `role.hire` and `role.fire` to the vocabulary, and row 199 adds `git.push.remote`. Under
REQ-WAIT-3 all three are hard waits, so today Jason resolves every hire, every fire and every push to a
remote. This version routes hire and fire for the chief of staff within fixed bounds, so the chief hires
and fires as needed and the user's per-agent forbid still holds. It routes `git.push.remote` for the coder
only (Mos, mos-1028 item 3). It adds agent host access with sudo through the credential store for
agents with a grant (Q183, Q184, Vikunja #424, due 2026-10-27). One `prd.approve` decision covers it all
(Vikunja #420).
When this version lands, a reviewed commit:
1. adds `role.hire`, `role.fire` and `git.push.remote` to `ROUTED` in
`packages/business/src/vocabulary.mjs` and in `packages/bus/src/broker.mjs`.
2. adds `role.hire` and `role.fire` to `roles/chief-of-staff.json` withinRole, and `git.push.remote` to
`roles/coder.json` withinRole. No other role file gets `git.push.remote`.
3. switches the broker's check on `instance.hire` and `instance.fire` from a cited decision to the
`reason` form, as `role.end` works for REQ-ROLE-2, and makes `publish`'s `decision` optional for a
within-role class (row 199 D9 step 1).
4. states, for a `git.push.remote` with no decision, where the source commit comes from: the latest work
commit the bus recorded for the coder's task. The check still matches it against the work ref tip and
the push sends it by oid (row 199, duckworth-0022 N9).
5. adds the host credential grant check and the host access it allows (REQ-CRED-3, Vikunja #424).
Every other check stays as rows 194 and 199 build it. Filling `hireable` stays a separate reviewed role
file commit in `roles/`.
## Text for the PRD
Status line, appended to the current paragraph:
> Version 1.7 routes the chief of staff's hire and fire actions and the coder's push to its own fork,
> and adds agent host access through the credential store, approved by Jason in decision
> <prd.approve decision id>.
Source, a new bullet above "Version 1.6 source":
> - Version 1.7 source, TARGET.md line 66: "Chief of staff agent has hire/fire authority for agents as
> needed; unless user forbids per agent". And Jason through Mos (mos-1067), DECISIONS Q183, in part: "I will
> still want to allow infra agent(s) on my .53 instance to access other servers, etc. However, not all
> agents will need to have that access. I envision a credential store within Mosaic Stack that works
> as some sort of variable for the agent to call. The system will enforce access to the credentials to
> keep the agents sandboxed with proper security. I need to be able to trust my agents with assisting
> me. This will require sudo access to several systems. If we can utilize the credential store in
> Mosaic Stack and allow agents to perform their jobs with NOPASSWD sudo on various machines, I am a
> happy man." DECISIONS Q184, on whether that is in place by 10-27: "Q5: by 10.27"
Under "Added in version 1.2", after REQ-COS-2:
> Added in version 1.7:
>
> - **REQ-COS-3.** The chief of staff hires and fires role instances without a decision each time. Hiring
> adds a role instance to the organization file and lists it for launch. Firing ends the instance's
> runs and removes it from the launch list. It stays in the organization file. Only the user deletes
> an instance. These actions are `role.hire` and `role.fire`. They leave the hard waits and route
> under this requirement, within these bounds:
> - The chief acts only in its own organization.
> - The chief hires only role definitions that need no credentials, until a later version adds
> credentialed roles.
> - The chief hires only the role kinds its role file lists. Changing that list is a reviewed commit.
> - The chief hires only a role whose every action is on a fixed list of plain work actions: sending
> messages, writing records, asking for review, updating its assigned tasks, and the scope and
> working-push actions of an assigned task. A role holding any other action, such as launch, end,
> organization, naming, hire, review verdict or merge authority, or any action added later, is never
> hired. So the chief never hires a chief, never picks its own reviewers or merge gate, and the author
> is never the reviewer. This holds whatever the role file's list says. Adding to that list is a PRD
> change.
> - The chief may set a hired instance's model and narrow its limits, and nothing else about it.
> - The user can forbid per agent. The user marks an instance so the chief can't fire it, end its runs
> or change its model, and marks role kinds the chief can't hire. Only the user changes these marks, through the
> WebUI or the CLI.
> - The kill switch stops every hire and fire while it is on.
> - Every hire and fire is logged with the chief's instance, its run and the reason.
Under "Landing", after REQ-LAND-1:
> Added in version 1.7:
>
> - **REQ-LAND-2.** A coder pushes its own work to its own fork. The action is `git.push.remote`.
> - What it pushes: one commit, by its oid, never a ref name. That commit is the latest work commit the
> bus recorded for the coder's task, and the coder made that push. A decision that authorizes a push
> names that commit, and it doesn't authorize any other.
> - Where it lands: only the branch `refs/heads/<its Gitea user>/<branch>`. The host builds that
> name, the agent never does. A push never forces.
> - Which remote: the agent's own fork of the project's upstream. The business file names the upstream,
> and the host never contacts it. For each push the host builds the fork's address over https from
> the upstream's host, port and repository name and the agent's own Gitea user. The history check,
> the push and the credential match all use that address. So the agent's own user owns the receiving
> repository on every push, and the agent never names it.
> - The host refuses an upstream on a host or port other than the agent's Gitea, and an upstream the
> agent's own user owns. It refuses a fork that is missing, empty or doesn't share the project's
> history. Gitea's push-to-create stays off, so a push never creates a repository.
> - Who may: the action routes under REQ-DEC-2 and is granted within-role to the coder role only. No
> other role file grants it.
> - Before version 1.7 the action was a hard wait under REQ-WAIT-3, so each push needed a decision
> the human resolved, and that decision named the commit.
>
> Every push is logged with the remote, the ref, the commit and, when one was given, the decision.
Under "Credentials", after REQ-CRED-2:
> Added in version 1.7:
>
> - **REQ-CRED-3.** An agent with a grant can run commands with sudo on another host. The agent asks
> for the host credential by name. The stack checks that a grant exists for that agent and that host
> before the credential is used. The agent never sees the key or any other secret value (REQ-CRED-1,
> REQ-VAR-2). An agent without a grant for that host is refused. Not every agent gets a grant. Who
> grants a host credential, and how, is set by the #424 design, which Jason approves.
REQ-WAIT-3, appended:
> `role.hire` and `role.fire` were added as hard waits and route under REQ-COS-3 since version 1.7.
> `git.push.remote` was added as a hard wait and routes under REQ-LAND-2 since version 1.7.
## Notes for the reviewer and for Mos
- REQ-COS-1 (1.2) already says the chief "hires and terminates role instances". REQ-COS-3 gives that
sentence its actions and bounds. It doesn't widen REQ-COS-1.
- The forbid now names what r3 called "replace": ending a kept instance's run with a reason and changing
its model (webby-0010 N2). A user-resolved decision still ends a kept run (row 194 D2.4).
- The fixed list is `HIRE_ALLOWED` in row 194 D2.3, an allowlist since r4 (webby-0010 B1). r3's text
named the refused authorities instead, which let merge-gate and any later action through.
- The kill switch is row 191's. If row 191 hasn't landed when this version is approved, the bound still
holds through row 194's chain on its candidate.
- REQ-LAND-2 follows row 199's design r6 (4d3007ec9 on megavolt/row199, approved by Duckworth in
duckworth-0027: D9 to D11, D9a, D9c, Q-M1 and Q-M2, and the oid binding from mos-1034 item 5) and
Megavolt's wording in megavolt-0005, megavolt-0008 and megavolt-0014, through Sage (sage-1377,
sage-1380, sage-1387). The fork bullet answers duckworth B5 and B6 and webby-0011 N1. Commit-email
checks, the refusal codes and the URL parse rules (D9c) are host checks with tests in row 199, not
bounds, so they stay out of the text (megavolt-0008).
- Once the routing lands and the decision is optional for the within-role class, the commit is the task's
latest recorded push at check time. It is still checked against the work ref tip and pushed by oid, so
the binding holds without a decision (megavolt-0008). Step 4 of "When this version lands" puts that in
the routing commit (duckworth-0022 N9, sage-1387).
- REQ-LAND-2 and REQ-COS-3 meet at the coder. Coder declares Gitea credentials, so REQ-COS-3's
credentials bound keeps the chief from hiring one until a later version lifts it. `git.push.remote` isn't
on REQ-COS-3's fixed list either, so a coder holding it stays unhireable until a PRD version adds it
(row 194 D2.3).
- Adding to the fixed action list is a PRD change (mos-1035 item 3, sage-1395). The role file's
`hireable` list stays a reviewed commit in `roles/`.
- The credentials bound names no row number, so the text doesn't go stale when row 199 lands. A later
version lifts it.
- REQ-CRED-3 states only what Jason said in Q183 and Q184 (sage-1422). It names the outcome and the
check, not how the stack reaches the host. Proxying the ssh and a signing socket are both designs, and
the #424 design picks one (duckworth-0038 B1).
- Open point: Q183 doesn't say who grants host access or how, so the text leaves it open and defines no
mechanism. The #424 design settles it, Jason approves that design, and Duckworth reviews it because it
gives agents sudo on hosts (Q184, Mos's reading, duckworth-0038 B2). The build is row 210 (#1707,
owner Gizmoduck), a new row on top of row 199 (megavolt-0019, sage-1430).
- This draft follows the PRD's own style: plain words, no semicolons in new sentences, and the status
paragraph never edited in place except to append.
-10
View File
@@ -1,10 +0,0 @@
# #436b build carry items
The design is approved at r5: 1be535aca, DESIGN.md sha256 68df02cdf890564c6ed0a5426204114389ba7eb0eea1cf04f1bf5938b6de45f4. Dewey gave the approval in dewey-0763 (verdict sha256 de52be1b…6b68). Section 7 is approved at fa5658941 by duckworth-0067 and is unchanged since. DESIGN.md stays as approved. The row's build takes the items below, and its BUILD.md cites this file.
- N9 (dewey-0763). Revoke writes `grant.revoked`, not a `grant.step`. A restart after a revoke on an open rotation therefore still sees a `commit` with no settle after it, and settles again: no file, `grant-lost`, a wrong message to Jason, and repeated deletes that come back 404 and go to `serviceLeft`. Two fixes:
- The section 5 restart row "`commit`, rotation, settle-pending" takes the held row's clause "with no `grant.revoked` after it".
- A 404 on a delete by id counts as done, not as a `serviceLeft` entry. That applies to revoke, sweep and the settle deletes.
- Add a test: revoke on an open rotation, then a restart. No settle runs, no Gitea call is made, and the restart sends no message about it.
- N10 (dewey-0763). `grant.revoked {serviceLeft}` can now hold two ids, so it is a list of ids everywhere. `sweep` reads it as one.
- duckworth-0067 N1 is in r4 (section 5, X-G15). Nothing more is carried for it.
-640
View File
@@ -1,640 +0,0 @@
# Gitea grant path, design draft r5 (Vikunja #436b, Gitea #1697 follow-up)
Author: Megavolt. Reviewer: Dewey. GitHub half: Bushroot. Base: Gizmoduck's #436a r6 (f9039682e, final,
`agents/gizmoduck/work/machine-grant/DESIGN.md` on origin/gizmoduck/machine-grant-design) and row 199's
DESIGN.md section 7a (e10e03f6b). Mos rulings used here: mos-1130 and mos-1136.
This draft builds nothing. It says how an Approve in Jason's inbox makes an agent's Gitea user, its token and
its forks with no manual step, and it answers mos-1130 item 3 (a to e) on the executor's admin credential.
## r5 changes (dewey-0761 on r4 3e66f5acf)
Section 7 is unchanged from fa5658941.
- Item 1 (blocks). A revoke on an open rotation deletes both tokens. When the last `grant.step` is a
rotation's `commit` with no settle after it (settle-pending or `grant-token-unknown`), section 6 step 3
deletes `previousTokenId` and the minted `tokenId`, each failure to `serviceLeft`. Section 5's
`grant-token-unknown` text now says both (X-G5).
- N8. Step 0 refuses `grant-token-unknown` while the agent is held, so a revoke is the only way out.
## r4 changes (dewey-0759 on r3 fa5658941, sage-1578, duckworth-0067)
Section 7 is unchanged from fa5658941, the text Duckworth approved and Gizmoduck cites.
- Item 1 (blocks), option (b), Sage's ruling in sage-1578. A step 0 refusal leaves the decision unspent only
until the next start. #436a r6 X3's start reconcile spends every resolved-yes grant decision with no
`action.allowed` as `interrupted`, and this draft no longer claims otherwise. `grants run` works while the
bus that refused is still up; after a restart the chief raises again and Jason approves again. `status`
says so on each open decision (D5, section 4 step 0, section 5, X-G11). (a) would change #436a r6, which is
final. The cost is in D5.
- Item 2 (blocks). A settle that gets 200 with an `id` that is neither token no longer leaves the file in use.
Memory drops the credential (199 `revoke`'s memory step), the file and both recorded tokens stay, nothing
is deleted at Gitea, and the step ends `grant.failed {code: 'grant-token-unknown'}` for Jason. A restart
keeps the credential out of memory (`Credentials` `held`, built in #436b's row like `reload`). Revoke clears
it (section 5, X-G6).
- N7. 3d says the private session-less writer runs `#secretCheck` on every guarded body, as `#event` does
(`broker.mjs:559`). The sentence is in 3d, so section 7 doesn't change.
- duckworth-0067 N1. Every reader of grant evidence filters on `kind = 'action.allowed'` and the grant actor
pair, so a `grant.*` name under another kind is not evidence (section 5, X-G15).
## r3 changes (dewey-0756 on r2 6db7cc45d, gizmoduck-0081 and -0084, duckworth-0065, mos-1144)
- B1 (blocks). Only `#consumeGrant`'s own record carries `decision`. Every other `grant.*` body names the
decision `grantDecision`, so `#spent` sees exactly one spend. A step 0 or step 1 refusal leaves the
decision unspent (section 7, X-G11, a mutant).
- B2 (blocks). Settle with the seat token is proven on scratch (F13): `GET /api/v1/token` answers 200 with the
minted id and 401 after the token is deleted. Only a 401 ends `grant-lost`. 403, 404, 5xx, a timeout, an
unreachable Gitea or a failed C1 to C4 leave the step open as `settle-pending`, delete nothing, and retry
on the next start and in `grants sweep` (section 5, X-G6). `Credentials.reload` is built by #436b's own
row in `packages/bus/src/credentials.mjs`; row 199's candidate doesn't change.
- B3 (blocks). Step 5 compares against the user's forks of every upstream in the agent's `publish` list, not
only `target.projects`. The rendered block names those other upstreams as reach the token may have. A fork
whose upstream has left `publish` refuses `gitea-reach-wider` with its name (section 2, step 5, X-G12).
- Section 7, the guard (gizmoduck-0081, -0084, duckworth-0065): one pattern, `grant.*`, on `body.action` and
on `body.operation` (Dewey N2: the consume record's `operation: 'grant.<method>'`). #436a renames its
`host.enroll.step`, `host.enrolled`, `host.checked` and `host.removed` into `grant.host.*`, so row 210's
`host.sudo` and `host.run` stay outside. The grant actor is `actor_role NULL, actor_run 'grant'`, which no
session or human event can produce (answers Dewey N1, same safety as `_bus`).
- D12 / Q-B4 closed by mos-1146 (section 8).
- N3. Step 5 reads every page of `/user/repos` and checks the total. N4. One nonce format, and the adopt
check compares the lowercased login. N5. `grants run` takes the `#inflight` claim. N6. One open-decision
query, `openGrantDecisions`, for `status`, rotate's void count and the start reconcile (section 7).
## r2 changes (from r1 714c07304, Dewey's dewey-0754, verdict sha256 20d63dfb…)
- D1 (blocks). `gitea_users` records the Gitea user `id`, and every later grant, rotation and fire compares it
(step 2). A nonce goes in `grant.step user` before the create and in the user's `full_name`, so a restart
adopts a user it made and refuses anything else (section 5, X-G6). Both are proven on scratch (F8, F9).
- D2 (blocks). The audit records are `action.allowed` bodies with reserved `grant.*` actions, plus a trigger
in the store migration that only `system/grant` writes them (section 7). A call and its result are two
records paired by `callId`, because `events` is append-only (3d). To settle once with Gizmoduck for #436a.
- D3 (blocks). Seat-token rotation gets its own rows. After `issue` has renamed the file, the executor settles
by asking Gitea which token the file holds (`GET /api/v1/token`, F7), and goes forward or back on that
answer, never by guessing (section 5).
- D4 (blocks). 3c and section 8 now state the real gap. Reading the admin file gives instance admin, which is
more than a forged Approve gives. `write:admin` resets any user's password, email and admin flag, proven on
scratch (F10). The separate uid is deferred on cost, and the draft says so.
- D5. The admin checks run before `#consumeGrant`, so a predictable refusal leaves Jason's Approve unspent. A
human `mosaic grants run <decision>` starts it again (step 0). The raise refuses when the admin slot is past
its rotate-by date. r4 (sage-1578): unspent only until the next start, when #436a r6 X3's reconcile spends
it as `interrupted`. The cost: if a step 0 refusal is followed by a restart before Jason rotates the admin
token or upgrades Gitea, the chief raises again and Jason approves again.
- D6. One list of start checks (3a, 3e). The "no organization" claim is now part of the bootstrap promise,
and the check is that `GET /user/repos` is empty, which catches org and collaborator repositories (F11, and
F12 for a site-admin token).
- D7. Admin rotation: a timestamped name (F9: names are unique per user), `--rotate-by` required, and `setAt`
changes, so open approvals are voided and `status` says so.
- D8. The restart's remote undo runs the same admin checks and writes the same call records.
- D9 (Q-B1). On fire: `prohibit_login` (F10) and delete every token the user has, not only the recorded one.
- D10 (Q-B2). Step 5 checks the seat token's real reach with `GET /user/repos` (F11). Section 8 says reach can
grow later.
- D11. One Gitea per business, stated, and checked at `set` and at raise.
- D12. Section 8 asks for the jason.woltje comparison. I can't check it myself.
## 0. What this adds to #436a
#436a section 11 left `method: "gitea"` for a later draft, because the executor needs an admin credential and
roles may never hold the admin scope (`packages/business/src/vocabulary.mjs:108-112`). That rule stays. The
admin credential here is not a role service: no `useFor` reaches it, no definition can list it, and only the
executor reads it.
Everything in #436a's sections 4 and 5 holds unless this draft says otherwise: the `credential.mint` raise
with a `grant` body, chief only (R3), the rendered block (R4), Jason's Approve via cli (R5, X1),
`#consumeGrant` with `operation: 'grant.gitea'`, the `#inflight` claim and abort (X5), `grant.step` records and
restart reconcile (X3), and revoke local first (section 6).
What row 199 already gives: the agent's `identity = {name, email, gitea: {url, user}}`, its
`publish: [project]` grant and each project's `remote` (199 D8, D9), the store and `issue(instance, service,
token, date, commit)` (`packages/bus/src/credentials.mjs:156`), and section 7a's rule that `issue` accepts a
resolved `credential.mint` decision as the human path. `issue` writes the file (tmp, fsync, rename), then calls
`commit`, then fills memory (`credentials.mjs:156-193`). Section 5 depends on that order.
## 1. Facts, proven on a scratch Gitea
Run on gitea/gitea:1.27.1 (sha256:fe4e4b1c8b93) on 127.0.0.1, throwaway data, container removed after each
run. Scripts and logs: `/mnt/storage/scratch/megavolt-r199/g436b/`. No log holds a token or password value.
F1 to F6 were reported in megavolt-0034, and Mos accepted them in mos-1136.
- F1 (`run4.log`). `POST /api/v1/users/{u}/tokens` with `Authorization: Basic <admin>:<admin token>` creates a
token for another user: 201. The same token in a `token` or `Bearer` header gets 401 on that route, while the
`token` header works on `GET /api/v1/user` (200). So the executor sends Basic on the token routes and
`token` everywhere else.
- F2 (`run4.log`). The child-scope cap: a child scope broader than the parent's gets 403 ("cannot create an
access token with a broader scope than the authenticating token"), for `write:organization` and for `all`.
A narrower one gets 201.
- F3 (`run4.log`). TOTP 2FA on the admin account blocks password Basic (401) and doesn't block token Basic
(201).
- F4. The route's check is `reqSelfOrAdmin`: whether the account is a site admin, not the token's scope. Any
`write:user` token on a site-admin account can mint a token for any user.
- F5. `write:X` implies `read:X` under Gitea's `HasScope`.
- F6. The cap is Gitea #38406, backported in #38426 (de4b8277e9 on release/v1.27, 2026-07-12). It first ships
in v1.27.0. v1.27.0-rc0 lacks it, and so does every 1.24.x, 1.25.x and 1.26.x through 1.26.4.
- F7 (`introspect.log`). `GET /api/v1/token` with the `token` header gives 200 with `id`, `name`, `scopes`,
`user`, `created_at` and `last_used_at`, and no token value. With Basic it gives 404. Scopes come back
normalized (3b).
- F8 (`r2probe.log` 1a-1d). `POST /admin/users` with a `full_name` gives 201 with the user's `id` and that
`full_name`. `GET /users/SEAT-ONE` finds `seat-one` with the same `id`: names match case-insensitively. A
create of `Seat-One` next to `seat-one` gets 422.
- F9 (`r2probe.log` 2a-2c). A second token with a name already used by that user gets 400. `GET
/users/{u}/tokens` with Basic lists the user's tokens.
- F10 (`r2probe.log` 4a-4f, 5a). With `write:admin`, `PATCH /admin/users/{u}` on another site admin sets a new
password (200; the new password then signs in, the old one gets 401) and a new email. On a plain user it sets
`admin: true`, then `admin: false` with `prohibit_login: true`. Without `write:admin` the same PATCH gets 403.
- F11 (`reach.log`). With the seat token (`write:repository`, `read:user`), `GET /user/repos` lists the user's
own fork, then also a repository it was made collaborator on, then also an org repository after it joined an
org team. `GET /user/orgs` and `GET /users/{u}/orgs` need `read:organization`, which neither token holds.
- F12 (`c4.log`). With a site-admin account's three-scope token, `GET /user/repos` is empty on an instance
where other users and an org it isn't in hold public and private repositories: being site admin doesn't add
them. It lists a repository once the account is a collaborator, and an org repository once it joins that
org's team. Joining an org with no repositories leaves it empty, and `GET /user/orgs` gets 403 (no
`read:organization`).
- F13 (`b2.log` b1-b7). Two seat tokens minted (201, ids 2 and 3). `GET /api/v1/token` with each token in the
`token` header gives 200, and the `id` is the minted one for both. After `DELETE` of seat-a's token (204), the
same call with it gives 401. After `prohibit_login` on seat-b (200), the call with seat-b's token gives 403
("This account is prohibited from signing in"), not 401. A value never issued gives 401. So 401 is the only
answer that means the token is gone; 403 means the account is shut, and section 5 leaves it `settle-pending`.
## 2. The binding
The shared part (section 10) is proposed to Bushroot in megavolt-0036. This section is the Gitea half.
```
grant: { "method": "gitea", "agent": "<instance>", "rotateBy": "YYYY-MM-DD",
"target": { "projects": ["<project>", ...] } }
```
Code fixes the scopes of the agent's token: `write:repository` and `read:user`, row 199 A2's set. The raise
can't name scopes, so the chief can't widen them.
Checks at raise, inside the raise transaction. Each refusal stores nothing:
- `agent` is a role instance of this business, its definition declares `{"service": "gitea"}`, and it has
`identity` and a `publish` grant (199 D8). Otherwise `gitea-not-declared`.
- Each project in `target.projects` is in the agent's `publish` list (`remote-not-granted`), and its
`remote`'s origin equals `identity.gitea.url`'s (`remote-host-mismatch`), the same checks as 199 D9 step 2.
- The business has an admin slot for `gitea` whose recorded `url` has the same origin
(`grant-admin-missing`), and the slot isn't past its rotate-by date (`grant-admin-rotation-due`).
- `rotateBy` is inside 199's rotation window.
The rendered block, appended by the broker as #436a R4 does: agent, Gitea user, email, Gitea URL, each fork
URL it will make (`<url>/<user>/<repo>`), the two fixed scopes, the rotate-by date, one of "new Gitea user",
"existing stack-made user, new token" or "token rotation" (step 2), and the line "The token can write every
repository this user can write. Step 5 checks that this is only its forks." Then, when the agent's `publish`
list has projects outside `target.projects`, the line "It may also reach this user's existing forks of:" and
each such upstream. The raise makes no network call, so the block names every such upstream, not only those
with a fork (B3). The 4096-character cap applies as in R4.
The raise event body gets `grant = {method, agent, target, rotateBy, definitionSha256, identitySha256,
adminSha256}`:
- `definitionSha256` as in #436a.
- `identitySha256` over the agent's `identity`, its `publish` list and the `remote` of every project in
`publish` (step 5 checks against all of them).
- `adminSha256` over the admin slot's metadata (`url`, `account`, `setAt`), never over the token.
At step 1 any of the three that differs refuses `decision-mismatch`. So a changed identity, publish list,
upstream or admin credential between raise and Approve voids the Approve (REQ-GRANT-2).
## 3. The executor's admin credential (mos-1130 item 3)
### 3a. A dedicated account
One Gitea account per Gitea instance, named `mosaic-grant` by default. It is a site admin and has TOTP 2FA on
(F3: the token path still works). Its password and TOTP secret stay with Jason and never reach the stack. It
is never jason.woltje or any other human's account. It owns no repositories, belongs to no organization and is
no repository's collaborator.
The admin checks, one list, run at `set` (3e), before every executor start (step 0) and before every remote
undo (section 5):
- C1. `GET /api/v1/version` is at least 1.27.0 (F6), compared as numbers, or `gitea-too-old`.
- C2. `GET /api/v1/user` gives `login == account` as recorded at `set`, and `is_admin` true, or
`grant-admin-account`.
- C3. `GET /api/v1/token` gives exactly the three scopes (3b), or `grant-admin-scope`.
- C4. `GET /api/v1/user/repos` is empty, or `grant-admin-not-dedicated`. By F11 and F12 this list covers
owned, collaborator and org-team repositories, and site admin alone doesn't fill it.
What code can't check: that nobody else uses the account, and org membership with no repositories (F11: the
orgs routes need `read:organization`, which the admin token doesn't hold and shouldn't). Both are the bootstrap
step's promise and go in the approval notes (section 8).
### 3b. The scopes, each shown needed
The admin token holds exactly `write:admin`, `write:user` and `write:repository` (mos-1136 item 1). From
`matrix.log`, with each scope dropped in turn:
| Parent token | A1 create user | A2 mint seat token | A3 fork (seat token) | A5 delete seat token |
|---|---|---|---|---|
| all four | 201 | 201 | 202 | 204 |
| without `write:admin` | **403** | 201 | 202 | 204 |
| without `write:user` | 201 | **403** | 202 | **403** |
| without `write:repository` | 201 | **403** (cap, F2) | 202 | 204 |
| without `read:user` | 201 | 201 | 202 | 204 |
- `write:admin` is needed for A1, and for `prohibit_login` on fire (F10).
- `write:user` is needed for A2 and A5.
- `write:repository` is needed only because of the cap. The seat's token holds `write:repository`, and a
parent must hold at least what it mints. The executor itself never writes a repository with it.
- `read:user` is covered by `write:user` (F5), so it is dropped.
C3 reads the scopes with the `token` header (F7). Gitea reports normalized scopes, so a token made with the
three plus `read:user` reads as the three. That is harmless, because `read:user` adds nothing to `write:user`.
A token made with `all`, or with another category, refuses.
### 3c. Where it lives, who can read it
- The token: `<configDir>/secrets/grants/<business>/gitea/admin`, 0600, in 0700 dirs, owned by the stack's
OS user. Written and checked by row 199's store code (D1: tmp, fsync, rename, the mode and owner checks).
- The metadata: `admin.json` next to it, 0600, holding `url`, `account`, `rotateBy` and `setAt`. No secret.
- A business grants on one Gitea. `set` refuses a second `url` while a slot exists (`grant-admin-exists`;
`remove` first), and the raise checks the agent's Gitea against the slot's `url` (section 2). Two businesses
on one Gitea each hold their own token for the same `mosaic-grant` account.
Who can read the file: the stack's OS user, and root.
Whether an agent process runs as that uid:
- On dragon-lin, yes. The stack runs as jwoltje, and so does every agent process on the machine: 13 claude, 4
codex and 2 pi processes, the bwrap sandboxes and node, counted 2026-10-20. That includes the T3 dev seats,
which are not sandboxed and can read any jwoltje file.
- Stack workers run in the seat sandbox, which has a private HOME and no `~/.config`, `~/.mosaic` or `~/.ssh`
(`packages/seat/src/sandbox.mjs`). A sandboxed worker can't reach the file by path. The sandbox doesn't
confine the network or the host uid (`sandbox.mjs:29-31`). So the file is safe from a sandboxed worker and
not from an unsandboxed process running as the same uid.
- On the .53 and .54 installs (#432) the stack runs as its own OS user. Workers run sandboxed as that user.
The T3 seats don't run there.
What a separate uid would buy. A same-uid process has two paths. It can forge a human proof (`verifyHuman` is
a cooperative same-UID check, `packages/bus/src/human.mjs:50-52`), and an approved grant then gives it a
two-scope token for a stack-made, non-admin user, because step 2 refuses foreign and admin users. Or it can
read the admin file, which gives instance admin (section 8). A separate uid for the file and the executor would
close the second path and leave the first. That is a real step up, from "a seat token for a stack user" to "no
admin".
I defer it on cost, not because it wouldn't help. It means the executor runs as its own process under its own
uid, with an IPC path from the bus that carries the decision id and nothing the bus could forge. That is a row
of its own, and it is cleaner after the human proof stops being same-uid (#435 behind Q175 sign-in). Jason
approves the gap as section 8 states it.
### 3d. Every use logged, and rotation
Every Gitea call made with the admin token, and every call the executor makes with a seat token, writes two
records (section 7 has where they live):
```
grant.call { callId, reason, method: "gitea", op } written before the call
grant.result { callId, status } written after it
```
- `op` is one of `version`, `self`, `token-scopes`, `repos-check`, `user-get`, `user-create`, `user-patch`,
`token-list`, `token-create`, `token-delete`, `admin-rotate`, `fork-get`, `fork`, `seat-self`,
`seat-repos`, `token-whoami`.
- `reason` is the decision id for a grant or its undo, the `grant.revoked` record id for a revoke, the
`grant.sweep` record id for a sweep, or `admin:<verb>` with the human actor for `set`, `rotate` and `remove`.
- No body, no header, no token, no URL query. `Credentials.assertClean` (`packages/bus/src/credentials.mjs:233`,
the runtime's `secretCheck`) runs on both. The private session-less writer of section 7 runs `#secretCheck`
on every guarded body it takes, `grant.step`, `grant.applied`, `grant.failed`, `grant.revoked`,
`grant.sweep` and the consume record included, as `#event` does (`packages/bus/src/broker.mjs:559`) (N7).
- `events` is append-only (`events_no_update`, `packages/bus/src/migrations/3b.sql:131`), so the result is its
own record, written in the transaction of the step's next `grant.step` where there is one. A `grant.call`
with no `grant.result` means the process died during the call. Reconcile lists each one in its message.
Rotation of the admin token: a human-only verb, `mosaic grants admin rotate gitea --rotate-by <date>`. It never
runs from an agent, the chief or the tester, and it doesn't need a decision because only `verifyHuman` can
start it:
1. C1 to C4 with the current token.
2. Mint a successor for the admin account itself: `POST /users/<account>/tokens` with Basic and the same three
scopes, named `mosaic-grant-<UTC yyyymmddThhmmssZ>`, so a second rotation on one day doesn't collide (F9).
F2's cap holds it to the current token's scopes.
3. C2 and C3 with the successor.
4. Write it through 199's tmp, fsync and rename. Update `admin.json` with the new `rotateBy` and a new `setAt`.
5. Delete the old token at Gitea by name, with the new token. If that fails, `status` shows
`adminTokenLeft: <name>` until `rotate --finish` succeeds.
A new `setAt` changes `adminSha256`, so every grant decision raised before the rotation and not yet run
refuses `decision-mismatch` (fail closed). `rotate` prints how many open grant decisions that voids, and
`status` lists them.
`status` shows the rotate-by date. Past it, the raise and the executor refuse `grant-admin-rotation-due` until
Jason rotates. A forgotten rotation stops grants rather than letting an old token run on.
If the admin token leaks, Jason deletes it in Gitea (Settings, Applications, as `mosaic-grant`) or disables
the account, then runs `mosaic grants admin remove gitea`, which unlinks the slot. Live agent tokens keep
working. Each one is revoked on its own (section 6).
### 3e. The one bootstrap step
Once per Gitea instance per stack, at #436b deploy. This is the only manual credential step left, and
Jason's approval names it:
1. In Gitea, Jason creates `mosaic-grant` as a site admin with no repositories and no organizations, turns on
TOTP, and as that user creates a token with `write:admin`, `write:user` and `write:repository`.
2. On the stack host, as the stack's OS user:
`mosaic grants admin set gitea --url <url> --account mosaic-grant --rotate-by <date>`, with the token on
stdin (never argv).
`set` is human-only (`verifyHuman`). It runs C1 to C4 (3a) before it writes anything, and any failure writes
nothing.
## 4. The executor's steps
0. **Start and pre-checks.** X1 starts the executor on Jason's yes. It claims `(agent, "gitea")` in
`#inflight` (#436a X5), then runs C1 to C4 and the admin rotate-by check. A refusal here writes
`grant.failed {code, grantDecision}` and leaves the decision unspent (section 7's spend rule). After the
fix (a rotation, an upgrade), Jason runs `mosaic grants run <decision>` (human-only, `verifyHuman`), which
starts step 0 again, claim included. That holds only while the bus that refused is up: at the next start,
#436a r6 X3's reconcile spends every resolved-yes grant decision with no `action.allowed` as
`interrupted`, a step 0 refusal included, and the chief raises again (D5, sage-1578). `status` lists each
such open decision with `spentAtNextStart: true`. While the agent is held (`grant-token-unknown`, section
5), step 0 refuses `grant-token-unknown`, so a revoke is the only way out (N8). While a start or another `grants run` holds the claim it refuses
`grant-busy` before any call, as #436a X2's `host-busy` does. It refuses `decision-consumed` for a spent
decision. No other path starts it. The tag is #436a X2 step 2's, over business, agent and `"gitea"`.
1. **Recheck and consume**, in one transaction: `#consumeGrant` with `operation: 'grant.gitea'`, the
definition still declares `gitea`, the three hashes match (section 2), no revoke of `(agent, gitea)` newer
than the raise. A failure rolls the transaction back, so the decision is not spent, and then writes
`grant.failed {code, grantDecision}`. Nothing remote has changed. `grants run` may retry it, but these
refusals (`decision-mismatch`, a newer revoke, the definition) recur until the chief raises again.
2. **User.** `GET /users/{user}` (F8: case-insensitive).
- 404: write `grant.step {step: 'user', nonce}`, where `nonce` is 16 random lowercase hex digits. Then
`POST /admin/users` with the agent's `gitea.user` and `email`, `full_name` = `<identity.name> [mosaic
<nonce>]`, a random password that is never stored, and `must_change_password: false`. Then, in one
transaction, record `gitea_users {user, id, nonce}` with the `id` from the response (F8) and
`grant.result`.
- 200, and `gitea_users` has this business and agent with the same `id`: go on. That is a second grant or a
rotation. A user deleted at Gitea and remade by a human under the same name has a new `id`, and refuses.
- 200 with any other `id`, or no `gitea_users` record: refuse `gitea-user-foreign`. The executor never mints
a token for a user this stack didn't create. Without this check an `identity` naming a human's account,
jason.woltje included, would get a token minted for that account on Approve.
- The user is a site admin, or has `prohibit_login`: refuse `gitea-user-admin` or `gitea-user-disabled`.
3. **Token.** Write `grant.step {step: 'mint', tokenName}`, with the name `mosaic-<tag>-<decision>`. Then
`POST /users/{user}/tokens` with Basic and scopes `write:repository` and `read:user`. The value is held in
memory only. Write `grant.step {step: 'minted', tokenId}`.
4. **Forks.** With the new token, for each project: `GET /repos/{user}/{repo}`. If it exists, its `parent`
must be the project's upstream, or `fork-foreign`. If it's 404, `POST /repos/{owner}/{repo}/forks`.
Push-to-create stays off on the Gitea side (199 N11).
5. **Verify.** With the new token: `GET /api/v1/user` gives `login == user` and the recorded `id`. Then
`GET /api/v1/user/repos?limit=50&page=N` from page 1 until a page holds fewer than 50, with the count
equal to `X-Total-Count`. More than 20 pages, or a count that disagrees, refuses `gitea-reach-unknown`.
Every listed repository must be owned by this user and have as `parent` the upstream of a project in the
agent's `publish` list (F11). Anything else refuses `gitea-reach-wider` and names it: a repository from a
collaborator or org-team add, or the user's old fork of an upstream that has left `publish`. So the token
never reaches past the rendered block (B3).
6. **Go live.** Write `grant.step {step: 'commit', previousTokenId}` (`null` for a first grant). Check the
abort signal, and with no `await` between, call 199 `issue(agent, "gitea", token, rotateBy, commit)`.
`commit` writes the `gitea_grants` record (tokenId, tokenName, projects, rotateBy, decision). The reference
is registered first (#436a X4).
7. **Old token.** For a rotation, `DELETE /users/{user}/tokens/{previousTokenId}` after the new one is live.
A failure adds `serviceLeft: <id>`, which `sweep` clears.
8. **Applied.** `grant.applied {grantDecision, agent, user, userId, tokenId, forks}` and a message to the human and
the chief. The token appears in no record, message or log.
The claim is released in a `finally`.
## 5. Failure and restart
Ordered so a failure leaves nothing an agent can use. "Delete the token" is `DELETE /users/{user}/tokens/{id
or name}` with the admin token, after C1 to C4, with `grant.call` and `grant.result` whose reason is the
decision (D8). Best effort; a failure adds `serviceLeft`.
**Settle.** Once `issue` has renamed the file, the file may hold the new token or the previous one, and the
record may name either. The executor and the restart never guess. They read the store file and ask Gitea which
token it is: `GET /api/v1/token` with that value in the `token` header (F7 for admin tokens, F13 for the
two-scope seat token; op `token-whoami`). The answer is an `id`, never a value.
- 200, `id` is the new token: forward. The `gitea_grants` record gets the new token (if `commit` hadn't
run), memory loads the file (`Credentials.reload(instance, service)`, a new method with load's checks,
built in #436b's row; row 199 doesn't change), and the previous token is deleted.
- 200, `id` is the previous token: back. The new token is deleted. The record keeps or gets the previous
token.
- 401: the file's token no longer exists at Gitea (F13 b4). The previous token can't be in the file, so the
agent holds nothing usable. Delete both by id, unregister, delete the record, unlink the file, and end
`grant.failed {code: 'grant-lost'}`. A new decision is needed. No file at all is the same case.
- 200, `id` is neither token: the file holds a live token this grant didn't mint and step 5 never checked.
Memory drops it (199 `revoke`'s memory step) and the agent gets `credential-unavailable`. The file, the
record and both recorded tokens stay, and nothing is deleted at Gitea. The step ends `grant.failed {code:
'grant-token-unknown', grantDecision, tokenId}`, where `tokenId` is the id Gitea gave, and Jason gets a
message. `status` shows `tokenUnknown`. At each start, before `Credentials` loads, the reconcile passes such
`(agent, "gitea")` pairs as `held` (a `Credentials` option built in #436b's row, like `reload`), so memory
never loads that file. A revoke (section 6) clears it: it unlinks the file and deletes both recorded tokens.
The unknown token is Jason's to look at.
- Anything else is `settle-pending`, and nothing is deleted: 403, 404, 5xx, a timeout, Gitea unreachable, or
C1 to C4 failing first. The last `grant.step` stays `commit`, the
`grant.call` without a `grant.result` stays listed, and `status` shows `settlePending`. The file stays as
it is, and the agent can use it: it holds one of the two tokens, and both passed step 5. The next start
and `mosaic grants sweep gitea` settle again. Only a 401 deletes.
| Fails at | State left | Cleanup |
|---|---|---|
| 0 | nothing, the decision unspent until the next start | none. `grants run` retries while this bus is up; after a restart the reconcile spends it (`interrupted`) |
| 1 | nothing | none |
| 2, before the create | nothing | none |
| 2, after the create, before the record | a user with the nonce | none here. Restart adopts it by the nonce (below) |
| 3, before `minted` | maybe a token nobody holds | delete the token by its recorded name |
| 4 or 5 | a token, maybe some forks | delete the token. Forks stay; they hold only upstream content |
| 6 before the rename, first grant | a token | unregister, delete the token |
| 6 after the rename, first grant | a store file with the new token, maybe the record | unlink the file, unregister, delete the record if written, delete the token |
| 6 before the rename, rotation | the previous grant intact | delete the new token |
| 6 after the rename, rotation | the file holds the new token, memory the previous | settle (forward: the new token passed step 5) |
| 7 | both tokens live, the new one in use | `serviceLeft: <previousTokenId>` |
**Restart**, before `Credentials` loads, from the last `grant.step`, as #436a X3. The local part runs first;
the remote part runs after the bus is up, with C1 to C4.
| Last step | Local | Remote |
|---|---|---|
| `user` | none | `GET /users/{user}`. 200 with `full_name` ending `[mosaic <nonce>]`, not admin: adopt, record `gitea_users` with its `id`. 404: none. Anything else: `gitea-user-orphan` with the name, for Jason to look at; nothing is deleted |
| `mint` or `minted` | none | delete the token by name |
| `commit`, first grant | unlink the file if present, delete the record if present | delete the new token |
| `commit`, rotation | none | settle |
| `commit`, rotation, settle-pending | none; memory loads the file as usual | none. The step stays open, and the next start or `grants sweep` settles again |
| `grant.failed {code: 'grant-token-unknown'}` with no `grant.revoked` after it | pass `(agent, "gitea")` as `held`; memory doesn't load the file | none. Jason's message repeats until a revoke |
Every reader of grant evidence (this reconcile, `status`, `openGrantDecisions`'s callers, `grants sweep`)
reads a `grant.*` record only where `kind = 'action.allowed'` and `actor_role IS NULL AND actor_run IS
'grant'`, so a `grant.*` name under another kind or actor is never evidence (duckworth-0067 N1).
Then `grant.failed {code: 'interrupted'}` (or `grant.applied` if a rotation settled forward) and a message to
Jason. A `grant.call` with no `grant.result` is listed in that message. After `interrupted`, the chief raises
again and Jason approves again.
## 6. Revoke
As #436a section 6, with the same revokers (the human, the assistant in its own organization, the tester on
dev only):
1. 199 `revoke`: memory, abort a use or the executor in flight, unlink the file.
2. One store transaction deletes the `gitea_grants` record and records `grant.revoked`.
3. After C1 to C4, `DELETE /users/{user}/tokens/{tokenId}` with the admin token. If Gitea can't be reached,
`grant.revoked {serviceLeft: <tokenId>}`, and `mosaic grants sweep gitea` clears it later. When the last
`grant.step` is a rotation's `commit` with no settle after it (settle-pending, or `grant-token-unknown`),
two tokens are live and the record may name either (section 5), so the revoke deletes both
`previousTokenId` and the minted `tokenId`, each failure listed in `serviceLeft` (dewey-0761). The id
`grant-token-unknown` named is not one this stack minted; it stays for Jason.
That closes 199 D12's gap: the token is gone at Gitea without Jason's A5 step.
The Gitea user and its forks stay after a revoke, so a new grant can reuse them.
On `credentials delete` (a fire), after the steps above and with the recorded `id` checked first:
- `GET /users/{user}/tokens` with Basic (F9), and delete every token it lists, not only the recorded one. A
token the record doesn't know about doesn't outlive a fire.
- `PATCH /admin/users/{user}` with `prohibit_login: true` (F10).
- Each step that fails is listed in `serviceLeft`, and `sweep` retries it.
## 7. Store records
Two tables, in #436a S1's migration if #436a row b and this row land together, otherwise a `3j-3k.sql`:
- `gitea_users`: business, agent, user, id, nonce, the decision that created it.
- `gitea_grants`: business, agent, user, tokenId, tokenName, projects, rotate-by date, decision.
The executor, revoke and `sweep` are the only writers, each in the transaction of its record. The admin slot
is a file, not a table, so a store backup holds no admin token.
The audit records (`grant.step`, `grant.call`, `grant.result`, `grant.applied`, `grant.failed`,
`grant.revoked`, `grant.sweep`) are `action.allowed` events with that name in `body.action`. They are not new
event kinds, so `events` and its CHECK (`3b.sql:9-30`) aren't rebuilt. This section is the rule's only home
(mos-1144). #436a r6 (f9039682e) uses it too: #436a row a's BUILD.md cites this section at the commit Duckworth
approves and maps r6's names to `body.action` values, with these renames (gizmoduck-0084):
`host.enroll.step` to `grant.host.enroll.step`, `host.enrolled` to `grant.host.enrolled`, `host.checked` to
`grant.host.checked`, `host.removed` to `grant.host.removed`. Cites are origin/refactor 00d4cc1f6.
**The guarded namespace** is one pattern: an `action.allowed` whose `body.action` or `body.operation` GLOBs
`'grant.*'`. `operation` is included because `#consumeGrant`'s consume record is
`{operation: 'grant.<method>', decision}` (#436a X1), and a forged one would spend a decision (Dewey N2). Row
210's `host.sudo` and `host.run`, written from agent sessions, stay outside it, and no grant-path name is left
under `host.*`.
**The grant actor** is `actor_role IS NULL AND actor_run IS 'grant'`. The only production insert into `events`
is `#event` (`packages/bus/src/broker.mjs:568`), which writes the session's role and run. Every session with a
NULL role also has a NULL run: the human (`:537`), the reader (`:549`) and a session-less record (`:1124`). The
tester's role is `tester`. So neither a session nor the human path can produce that pair, as no instance can
be named `_bus` (`:166-168`). A `system` role would not be safe, because `system` matches `ID_PATTERN`
(Dewey N1). The broker writes guarded events, the consume record included, through a private session-less
writer added in #436a row a; `#event` is unchanged. Revoke, sweep and check run for the human, the assistant,
the tester or the chief, and their events still go out under the grant actor, with `body.by` naming the caller
the broker checked.
**Two guards** stop a forged record:
- `recordEvent`, the trusted adapters' path (`broker.mjs:1111`), refuses `reserved-event` for any
`action.allowed` in the guarded namespace, as it does today for `git.push.working` and the merge action.
- The migration adds a trigger: an `action.allowed` in the guarded namespace aborts unless
`actor_role IS NULL AND actor_run IS 'grant'`. That covers any path that misses the first guard.
**Spend rule (B1).** `#spent` (`broker.mjs:811-817`) counts any `action.allowed` that names the decision in
`body.decision` and whose `operation` isn't `decision.raise`; a record with no `operation` counts. So the
consume record is the only guarded body with a `decision` key. Every other `grant.*` body names the decision
`grantDecision`. A step 0 or step 1 refusal therefore leaves the decision unspent, and the start reconcile
for "resolved yes, unspent" keeps reading `#spent` unchanged.
A wider gap sits under this one and isn't this row's: today any `recordEvent` `action.allowed` that carries
`body.decision` spends that decision. Follow-up for Sage to file.
**Open decisions (N6).** One query, `openGrantDecisions(business, method)`: `credential.mint` decisions
resolved yes by the human whose raise body has `grant.method = method`, and for which `#spent` is false.
`status` lists them, `grants admin rotate` counts the ones it voids (all of them, since `adminSha256`
changes), and the start reconcile reads the same query.
**No task refs.** Guarded bodies never carry `target` or `task_ref` holding a task ref, so they stay clear of
`task-subject-required` (`broker.mjs:561-565`) on any path. No other `action.allowed` reader sees these
events: the rest key on `decision.raise` (`:670`, `:803`), `git.push.working` (`:863`, `:2019`) or
routine (`:2100-2102`).
## 8. Approval notes for Jason
Stated once, here, for the #436b approval:
- **Site admin plus `write:user`, and `write:admin`.** The admin token can mint a token for any Gitea user,
with any scope up to its own three, including users the stack didn't create (F4). With `write:admin` it can
also set any user's password, email and admin flag, and disable any user (F10, proven on scratch against a
second site admin). Whoever holds the token can take over any account on that Gitea, jason.woltje's
included, make any user a site admin, and read or write any repository. The executor uses none of that
beyond its steps, but the token itself can.
- **Same uid on dragon-lin.** The token is a file readable by jwoltje, and every agent process on dragon-lin,
T3 dev seats included, runs as jwoltje (3c). The seat sandbox keeps sandboxed workers away from it. It does
not keep an unsandboxed same-uid process away. Such a process gets instance admin by reading the file, which
is more than it could get by forging an Approve. A separate uid would close that path. It is deferred on cost
(3c), not because it wouldn't help.
- **For comparison, today (D12).** Every dev seat on dragon-lin already pushes as jason.woltje through
`scripts/git-credential-mosaic.mjs` (mos-1106). jason.woltje is a site admin on git.mosaicstack.dev, with
restricted false (mos-1146: Mos read `/api/v1/users/jason.woltje` with the mos token at 23:05:29Z; Jason's
token wasn't used). So a same-uid process on dragon-lin can already act as a site admin there through that
credential, and the admin file adds little exposure on that Gitea beyond what the seats already have. It
still matters on any other Gitea a business grants on, and once seats stop pushing as jason.woltje (#431).
- **A seat token's reach can grow.** Step 5 checks, at the grant, that the token reaches only its forks. If
someone later adds the user to an org or as a collaborator, the token reaches that too, until the next grant
or rotation checks again.
- **The admin account is dedicated by promise.** Code checks that it is a site admin with these three scopes
and no reachable repositories (C1 to C4). It can't check that nobody else uses it or that it is in no org.
- **The bootstrap step** (3e) is the one manual credential step: Jason creates `mosaic-grant`, its TOTP and
its token, and enters the token once on stdin.
## 9. Tests and mutants (for the row, not built here)
Unit tests use a fake Gitea over TLS, as row 199's publish tests do. Section 1's facts get a scratch
gitea/gitea:1.27.1 test on 127.0.0.1 with throwaway data, removed after. No test contacts git.mosaicstack.dev
or any other live service. No token value is printed.
| Test | Claim |
|---|---|
| X-G1 | a gitea raise for an agent without `gitea`, `identity` or `publish`, or a project outside `publish`, a remote on another host, no admin slot, or a slot past rotate-by refuses and stores nothing |
| X-G2 | `set` refuses a version below 1.27.0 (fake reports 1.26.4, 1.27.0-rc0, and 1.9.0 against 1.27.0 as a string trap), a non-admin, a login mismatch, an account that reaches a repository, a token with another scope or `all`, and a second `url`. It writes nothing on any refusal |
| X-G3 | on scratch 1.27.1, end to end: user made with the nonce and its `id` recorded, token minted with the two scopes, forks made, step 5 passes, `issue` writes the 0600 file, `grant.applied`, then revoke deletes the token at Gitea (gone from its list) |
| X-G4 | an existing user with no record, a recorded name whose `id` differs (deleted and remade, and a case variant of the name), a site-admin user and a disabled user each refuse before any token call |
| X-G5 | each failure row of section 5 leaves the stated state, rotation rows included. A failed delete leaves `serviceLeft`. A revoke while a rotation's `commit` is settle-pending, and one after `grant-token-unknown`, deletes both `previousTokenId` and the minted `tokenId` at the fake, and a failure of either lands in `serviceLeft`. A new grant for a held agent refuses `grant-token-unknown` at step 0, before any Gitea call |
| X-G6 | a restart after `user` adopts a user whose `full_name` ends with ` [mosaic <nonce>]` (16 lowercase hex), whose lowercased login equals `gitea.user` lowercased and who isn't an admin; does nothing on 404; and reports `gitea-user-orphan` for a different `full_name`, a different login or an admin. A restart after `mint`, `minted` and `commit` of a first grant leaves no token at Gitea, no file and no record. A restart after `commit` of a rotation settles forward when the file holds the new token and back when it holds the previous one, ends `grant-lost` only on a 401, and stays `settle-pending` with nothing deleted on 403, 404, 500, a timeout and an unreachable fake. On 200 with an unknown `id` it drops the credential from memory, deletes nothing at Gitea, keeps the file, ends `grant-token-unknown`, and a second restart still doesn't load the file; a revoke then clears it |
| X-G7 | an identity, publish, upstream or admin slot change between raise and Approve refuses `decision-mismatch` before any Gitea call. An admin rotate voids open grant decisions and `status` lists them |
| X-G8 | every admin-token and seat-token call writes `grant.call` before it and `grant.result` after it, paired by `callId`. A crash between them is listed by reconcile. No record body holds a token |
| X-G9 | admin rotate: C1 to C4, the successor named by time, checked, written, the old token deleted. Two rotations in one second's name refuse cleanly, not half done. Past rotate-by, the raise and the executor refuse |
| X-G10 | seat-token rotation: the new token is live before the old one is deleted |
| X-G11 | a step 0 refusal (rotation due, too old) and a step 1 refusal leave `#spent` false, and `grants run` by the human gets past `#consumeGrant`. `grants run` from an agent, the chief or the tester refuses. A `grants run` while a start holds the claim refuses `grant-busy` before any Gitea call. No `grant.*` body other than the consume record has a `decision` key. A step 0 refusal, then a restart, then `grants run`: the reconcile has spent the decision as `interrupted` and `grants run` refuses `decision-consumed`; before the restart `status` shows `spentAtNextStart` |
| X-G12 | step 5 refuses `gitea-reach-wider` when the user was made collaborator on another repository, when it joined an org team, and when it owns a fork of an upstream that has left `publish`, and deletes the token. It passes with an old fork of a `publish` project outside `target.projects`, which the rendered block named. A fake returning 51 forks over two pages is read in full; a total that disagrees refuses `gitea-reach-unknown` |
| X-G13 | `recordEvent` refuses an `action.allowed` with `body.action` `grant.call` (and each `grant.*`) and one with `body.operation` `grant.gitea`. A direct store insert aborts on the trigger for `grant.step` and `grant.host.enrolled` from a session, and for `grant.step` under the human's NULL pair. `host.sudo` from an agent session still writes |
| X-G14 | fire: every token the user has is deleted, the user gets `prohibit_login`, and a failure of either lands in `serviceLeft` |
| X-G15 | a `grant.step {step: 'commit'}` stored under `action.refused`, and one under the human's NULL pair, are ignored by the restart reconcile: no settle runs and no Gitea call is made |
Mutants, a first list: drop the version floor; compare the version as a string; accept another admin scope;
skip `gitea-user-foreign`; compare the user by name instead of `id`; skip the admin-user refusal; adopt a user
at restart without the nonce; create the user before writing `grant.step user`; mint before writing
`grant.step mint`; delete the old token before `issue`; settle a rotation by the record instead of
`token-whoami`; run C1 to C4 after `#consumeGrant` instead of before; drop the `identitySha256` compare; drop
`adminSha256`; skip the `grant.call` write; write `grant.result` before the call; drop the `grant.` reservation
in `recordEvent`; drop the trigger; widen the trigger to `host.*` (breaks `host.sudo`); guard `body.action` only, not
`operation`; key the trigger on `actor_run` alone; put `decision` in step 0's `grant.failed`; delete both
tokens on a settle 403 or 5xx; compare step 5 against `target.projects` instead of `publish`; read one page
of `/user/repos`; let `grants run` skip the claim; send the admin token in a `token` header on the mint route; revoke the
remote token before the local steps; skip the fork `parent` check; skip step 5's reach check; delete only the
recorded token on fire; let `grants run` accept an agent; keep the file in use on a settle 200 with an
unknown `id`; load a `held` pair at start; revoke only `tokenId` on an open rotation; let step 0 pass a held agent; drop the `kind` filter from the restart reconcile.
## 10. Shared with Bushroot's GitHub half (proposed in megavolt-0036, answer pending)
- S1. The `grant` body: `{method, agent, rotateBy, target}`, scopes fixed by code per method.
- S2. The admin slot: `<configDir>/secrets/grants/<business>/<method>/admin`, set once by
`mosaic grants admin set <method>`, never an agent service. One service instance per method per business.
- S3. `grant.call` and `grant.result` (r2: two records, paired by `callId`) for every service call, as reserved
`action.allowed` bodies (section 7).
- S4. The executor shape: pre-checks before consume (step 0), claim, consume, mint, verify, `issue`, applied,
with `grant.step` before and after the mint, and settle by asking the service which credential the file
holds. GitHub may not fit "mint once and store" (App installation tokens expire within an hour). If so, S4
becomes per method and the rest stays shared.
- S5. Revoke local first, then the record, then the service, with `serviceLeft` and `grants sweep`.
- S6. A capability floor per method, checked at `set`, before every start and before every remote undo.
Dewey notes items 2, 3 and 5 of dewey-0754 apply to S3, S4 and S6. The r2 text above carries them.
## 11. Questions
- Q-B1. Answered (dewey-0754 item 9): on fire, `prohibit_login` and delete every token (section 6). The PATCH
route is proven (F10).
- Q-B2. Answered (dewey-0754 item 10): no organizations. The token's real reach is checked at step 5 and
stated in section 8.
- Q-B3. Bushroot's answer on S1 to S6.
- Q-B4. Answered (mos-1146): jason.woltje is a site admin on git.mosaicstack.dev, restricted false. Section 8
says so.
- Q-B5. Answered (gizmoduck-0081, -0084, mos-1144): one rule for #436a and #436b, section 7. #436a row a's
BUILD.md cites section 7 at the commit Duckworth approves, or carries its text in full if row a's candidate
comes first. Duckworth checks it with that candidate.
-167
View File
@@ -1,167 +0,0 @@
# Row 199 (#1697): per-agent credential store, each agent's own git identity, publish to its fork
Builder: Megavolt. Brief: `docs/plans/2026-10-20_agent-credentials.md`. Design: `DESIGN.md` r6 at
4d3007ec9, approved by Duckworth in duckworth-0027 (REVIEW-r6.md at 640011518). Authority model ruled
by Mos in mos-1027. Branch: megavolt/row199. Reviewer: Duckworth.
Status: in progress. The full gate has not run; it is Sage's. Only single test files have run on
this branch, apart from one slip noted under "Runs".
## Commits so far
- 286bf598c business: agent identity, the publish grant, the project upstream with the D9c parse, store
credential references; git.push.remote as a hard wait.
- a2c387dba bus: broker publish authority (checkPublish, recordPublish, publishRefused, pushCommit on
the raise); the D9c parse moves to `business/src/vocabulary.mjs`; pm.md and chief-of-staff.md list
git.push.remote among the hard waits.
- 609dc9364 bus: the per-agent credential store in `credentials.mjs`.
- 504336797 bus: runtime wiring, the store root and the four publish entries in the op table.
- a18fc6cf8 tasks: request calls use `useFor` keyed from the session; the start probe keeps `use`. The
D5 test in `credential-store.test.mjs` pins the `use(` call sites to `credentials.mjs` (useFor's own
call) and `tasks/src/adapter.mjs` (the host's @sync and start probe).
- aa740cc71 bus: the lifecycle verbs `credentials.issue`, `credentials.revoke` and `credentials.delete`
in `credential-verbs.mjs`, the broker branch, the runtime wiring; `organizationWriter` gets a
`detail` on `write` and a receipt-only `record`.
- 0cdace415 cli: `mosaic credentials status|issue|rotate|revoke|delete`, the scripts/mosaic route, docs.
## Deviations from DESIGN.md r6
1. **Identity is optional on an agent.** D8 reads as identity on every agent. Today's business files
carry none, so identity stays optional and `publish` requires it (N10's load refusal). An agent
without identity commits as E7 (T6).
2. **Revoke of a `{file}` or `{env}` reference.** Section 6 describes revoke for store references.
For a reference outside the store, revoke forgets the token in memory and never unlinks the file;
the lifecycle verb then turns the business file's reference into `store`, so a restart finds
nothing. The test: `credential-store.test.mjs`, "revoke of a {file} reference".
3. **Issue order.** Issue writes the token file (tmp, fsync, rename, dir fsync), then runs the
caller's commit (the business file's reference becomes `store`), then sets memory. A failed commit
leaves the file on disk and the token out of memory, so the agent stays `not-issued` until the
issue is run again. The test: "a failed commit leaves the token on disk but not in memory".
4. **The D9c parse lives in `business/src/vocabulary.mjs`, not `business.mjs`.** The broker reads
`forkProblem` and `forkUrl`, and `project-runtime.test.mjs` pins the broker's import closure to
files that import nothing. `business.mjs` wraps `parseGiteaUrl` with the loader's refusal
messages. The pin test names the two new imports.
5. **`withGitea` takes the publish ticket, not a cap (D9b).** The key is the session checkPublish
checked with `#agent` (a run holding its claim; never a tester, reader or human), read through
`broker.publishSession(ticket)`. The token is usable only between checkPublish and recordPublish
or publishRefused. Its refusals are filed once, by publishRefused on the task's trail, and not
also by `refusing()`. The test: `publish.test.mjs`, "runtime: withGitea keys from the open ticket
only".
6. **Role contracts.** `workflow-roles.test.mjs` requires the hard waits listed in `roles/pm.md` and
`roles/chief-of-staff.md` to equal GATED_ONLY. Adding git.push.remote to the vocabulary made it
fail, so both lists now name git.push.remote. Prose only; no hash pin reads either file. Duckworth
should review this hunk because `roles/` is reviewed authority.
7. **Delete takes a reason.** Section 6 lists `delete AGENT` with no reason. The broker requires
`reason` on delete as on revoke, so every receipt that ends a credential says why.
8. **Rotate is `credentials.issue`.** There are three bus verbs, not four. An issue over an entry the
bus holds is recorded as `rotate` in the receipt and the answer. The CLI's `rotate` reads the
status first and refuses before reading stdin when nothing is issued.
9. **Issue covers entries the bus loaded.** The business file must already name the kind for the
agent, and the bus must have loaded that reference at start; otherwise `credential-unknown`. An
agent added after the bus started needs the restart row 165 already asks for. Row 194's human
`create` is where a new agent's `store` references come from.
10. **No Console page yet.** Section 6 puts the verbs in the Console's agent page too. This branch has
the bus verbs and the CLI. The Console page is a separate piece (packages/webui, its `BUS_VERBS`
and `STATUS` pins); I raise the split with Sage before the manifest. Ruled in sage-1421: the
Console page is out of row 199 and becomes a follow-up row with Webby as reviewer. The manifest
holds bus, business, tasks, cli and seat.
11. **The push runs in the bus child, and the git socket polls for it (D9b).** The git socket runs in
the host process. The broker, the runtime and the token are in the bus child that
`packages/cli/src/host.mjs` forks, and host ops cross that IPC one at a time with a 10 s answer
limit. ls-remote and the push can run longer than that, so `withGitea(ticket, fn)` can't wrap a
host-side git. The split: the socket runs D9 steps 3 and 4 (tip and identity under the lock,
`working.publish`), then op `publishPush`, which starts `publishRemote` in the bus child under
`withGitea` and answers `{state: 'running'}`. The socket polls op `publishResult` every 200 ms
until it reads `pushed` or `refused`. Fork, user, source and destination come from the ticket
(`broker.publishTarget`), never from the host's message. `recordPublish` also refuses `ref-moved`
when the pushed ref isn't the ticket's destination. A second `publishPush` on a running ticket
refuses `publish-busy`. The token reaches git on fd 3 only (`runGit` option `pass`), read once
by a one-shot credential helper that answers `get` for https on the fork's exact host and port.
On close, a running publish's git isn't awaited; git's own time limit ends it, and the ticket
goes with the process. The 10 s IPC limit stays as it is (sage-1440). The poll has a bound
(sage-1440): `publishDeadlineMs`, by default four times git's limit and grace (publishRemote runs
at most four git commands) plus 30 s. Past it the socket refuses `publish-timeout`, filed once by
publishRefused, and logs that the bus process's push may still reach the fork. A failure that is not a refusal is filed as `git-failed` and logged on
the bus's stderr by error name and code only. Tests: `seat/tests/publish.test.mjs` (11),
`bus/tests/publish.test.mjs` (the process ops), `cli/tests/publish-socket.test.mjs` (7, end to
end through the socket, the in-process runtime, the store and a TLS fake Gitea; one is the
publish-timeout outcome).
12. **No harness tool for publish.** `packages/harness/src/git-tools.mjs` has push, fetch and land
clients. A runner tool that calls the socket's `publish` op is in packages/harness, outside the
manifest scope sage-1421 set. Today a publish needs a socket client; the tool is a follow-up.
13. **The bind rule's `git.push.remote` clause has no test of its own.** D11 adds it and the
launcher has it. A role can publish only a push its own role made (`work-not-yours`), and
`recordPush` needs `git.push.working`, which already binds the socket. No role reaches the
clause alone, so dropping it changes no outcome today. It matters once a role can publish work
it didn't push. Ruled in sage-1440: keep it; its mutant goes in the section 11 list as
equivalent, with this reasoning, and Duckworth judges it.
14. **A push between checkPublish and the git push refuses; it never lands a different commit.**
T3 says such a push "doesn't change the commit that lands". Under the hard wait it refuses:
`working.publish` checks the work ref's tip against the ticket's source under the lock
(`ref-moved`, `seat/tests/publish.test.mjs`), and `recordPublish` checks `pushCommit` again
(`decision-mismatch`, `bus/tests/publish.test.mjs` "r4 B4"). The fork gets only the ticket's
oid either way. N9's routing-commit test still applies. Ruled in sage-1440: T3's wording in
DESIGN.md now reads "refuses with `ref-moved` or `decision-mismatch`", and an end-to-end test
asserts each code mid-flight in `cli/tests/publish-socket.test.mjs`: a push after checkPublish
refuses `ref-moved` with no request to the fork; a push recorded while the bus process pushes
refuses `decision-mismatch`, filed once, and the fork holds only the ticket's oid.
15. **Fix: rotate on a due gitea key (gizmoduck-0050, sage-1494).** The CLI's `ISSUED` set
(`packages/cli/src/credentials.mjs`) listed valid, expiring and expired, but the bus's `status()`
gives a gitea key past its rotate-by date `rotation-due`. So `credentials rotate` refused the one
case it exists for. `ISSUED` now names every state status() gives a held token. The test:
`cli/tests/credentials.test.mjs`, "rotate goes ahead on each held state". Its mutant (the old set)
fails that test and goes in the section 11 list.
## N9, for the routing commit (sage-1388)
Today git.push.remote is a hard wait, so every publish carries a decision, and the decision's
`pushCommit` fixes the commit. If a later commit routes git.push.remote within role, a publish has
no decision and no `pushCommit`. `#publishFacts` then takes the commit from `#latestPush`, and the
git socket's tip check under the lock is the only thing that holds it. That commit needs a test
that a push between checkPublish and the git push doesn't change what lands.
## R1, the backup set
Checked at 504336797. Row 185's `backup_store` (`scripts/stack-release.sh:372-392`) copies only
`<dataRoot>/bus/bus.sqlite`, plus its sums, modes, listing and stamps, into
`<dataRoot>/backups/stack/<activation>/`. Nothing under `<configDir>` is copied, so
`<configDir>/secrets/agents` isn't in the backup set. `rehearse` copies the live business file
through `R_BUSINESS_JS` (`:1268-1288`). Its `swap` replaces every credential reference, a `store`
one included, with a placeholder file in the rehearsal's own dir. The guard then refuses
`credential-ref-live` unless every reference is such a placeholder (`:1329-1337`), so a rehearsal
never reads a store token. A backup tool outside this repository that copies `<configDir>` would
copy tokens. That is R1's open half, and it is Jason's to check on each VM.
## T1 limit
A wrong owner on a store dir or file can't be set up without root, so T1 covers modes, symlinks and
dirs. The owner check (`st.uid !== process.getuid()`) is in `privateDir` and `#read`, and is untested.
## Runs
Single files, each green on its commit: bus broker, decision-close, decision-route, decisions,
hard-waits, land, project-runtime, launch-notice, review, publish, credential-store, credentials,
credential-status, limits, tester, plugins-verb, routes-verb, limits-runtime, instances-verb, web,
usage-runtime, limit-checks, daily-quota-runtime, chief-of-staff, tasks, tester-writes; every
business test file. Since: credential-verbs (2/2), credential-store (11/11, with D5), each tasks file,
tester, credential-status, plugins-verb, routes-verb, instances-verb; business launch-settings,
plugin-settings, instance-settings, token-limit-settings; cli credentials (3/3), cli, app,
tester-reads, tester, other-tree, transport, instances, limits, verbs. Hash pins: none on a changed
file (`thread-migration.test.mjs:46` pins DDL only).
Publish, single files, each green before its commit: seat
checkout (77), working (26), publish (11), landing (8); bus process (8), land (17), publish (12),
credential-verbs (2); cli publish-socket (4), git-socket (24), launcher (29); harness tools (22).
Hash pins: none on a changed file.
sage-1440 changes (the poll bound, the mid-flight tests), single files, green before the commit: cli
publish-socket (7), git-socket (24), launcher (29). Hash pins: none on a changed file.
The rotation-due fix, single files: cli credentials (4), green; the old-set mutant fails it 3/1.
Hash pins: none on a changed file.
Slip, 2026-10-20 about 19:51Z: I ran `node --test packages/tasks/tests/*.test.mjs` in one command,
which is a package suite, without asking Sage for the slot. It passed, 97 pass and 0 fail, in under a
minute. Reported to Sage in megavolt-0017.
-430
View File
@@ -1,430 +0,0 @@
# Row 199 design: a credential store per agent, own git identity, pushes to a remote
Megavolt, 2026-10-20. Gitea #1697, MVP T1 item 6. Brief: `docs/plans/2026-10-20_agent-credentials.md`.
Reviewer Duckworth. Mos rules on the authority model (section 9) before any code. Base: refactor at 992650232.
No credential value was read to write this. Evidence is code and docs only.
r2 answers duckworth-0013 (review at 6ab9cd927 on duckworth/reviews): B1 in D3a, T1 and T5, B2 in D9 step 5
and T3, N1 in section 6 and Q-M3, N2 in section 6, N3 in D9 step 3, N4 in D9a and T3, N5 in E5 and D9b.
r3 records Mos's rulings (mos-1027) in section 9. The Q-M2 hard wait changes D9 step 1, D9b and T3. Q-M1
drops O2 (D10, A3, R2).
r4 answers duckworth-0017 (REVIEW-r3.md at 4065ed59f): B3 in T3 and D9 step 5, B4 in D9 steps 1, 3 and 5, N6
and N7 in D9a, N8 in D9 step 1.
r5 takes webby-0011 N1 (row 194 review, through honker-0009): the host checks at each push that the
remote's owner is the agent's own Gitea user. r4 named one `remote` per project, which can't be a fork per
agent (O1) once two agents publish the same project. So `projects.<p>.remote` now names the upstream
repository, and the host builds each agent's fork URL from it (D9 step 2, D10, T3). It also fixes a conflict
found while building: D8 made `identity` required for every agent whose definition declares `gitea`, while D9
step 2 says today's files load unchanged. Coder declares `gitea`, so r5 makes `identity` optional and
`publish` require it (D8).
r6 answers duckworth-0026 (REVIEW-r5.md at 3cb6888e1), with Mos's acceptance (mos-1043). B5: the derived URL
has one name, `fork`, and D9a's `ls-remote`, step 5's push, the helper's host match, `checkPublish`'s result
and step 6's event all use it (D9, D9a, D9b, T3). B6: the URL parse is written down (D9c) and tested (T3).
N10 in D8 and T3. N11 in D9a and A3. N9 is carried to BUILD.md for the routing commit.
## 0. Terms
- **Stack instance**: one Mosaic Stack deployment (row 190). It has one OS user, one config dir
`~/.config/mosaic-NAME` (here `<configDir>`), one data root and one bus. Case 1 and Case 2 are two stack
instances on two VMs (DECISIONS Q180).
- **Agent**: a role instance in a business file (`roles.<agent>`). Today's code calls it "instance". This
design says agent.
- **Kind**: a credential type: `gitea`, `vikunja`, `model`.
- **Host**: the bus process and its git socket, running as the stack instance's OS user, outside any sandbox.
## 1. What exists (measured at 992650232)
| Code | Fact | Where |
|---|---|---|
| E1 | The business file references credentials per agent: `roles.<agent>.credentials.<kind> = {file or env, rotateBy or expires}`. The kinds an agent may hold are the ones its role definition lists | `business/src/business.mjs:89-111`, `credentials.mjs:18-38`, `role.mjs:105-118` |
| E2 | The bus reads each token once at start (O_NOFOLLOW, own uid, 0600, at most 16 KiB, outside the repos and data root) into a private map keyed `<business>/<agent>:<kind>` | `bus/src/credentials.mjs:18-89`, `bus/src/runtime.mjs:78-89` |
| E3 | Only Vikunja uses a token: `credentials.use("<business>/<agent>", "vikunja", ...)`. Gitea tokens are loaded and never used. `assertClean` refuses any event that contains a token | `tasks/src/adapter.mjs:70`, `runtime.mjs:134`, `broker.mjs:560` |
| E4 | Agents run in bwrap as the bus's own uid. `~/.config`, `~/.mosaic` and `~/.ssh` are not bound. The git socket is bound only for `git.push.working`, `review.verdict` or a landing | `seat/src/sandbox.mjs:21-30, 104-124`, `cli/src/launcher.mjs:401` |
| E5 | A session proves who it is with a 64-hex cap the broker mints and maps to a frozen session record. The git socket resolves the cap to (business, role) before anything else | `broker.mjs:280, 379-383`, `packages/cli/src/git-socket.mjs:398-407` |
| E6 | The model credential is one env file per host (`EnvironmentFile=-%h/.config/mosaic-dev/secrets/claude-session.env`). `ENV_ALLOW` passes it into every sandbox, so every agent holds the same model token | `cli/systemd/mosaic-bus.service.in:24`, `seat/src/session.mjs:83-94` |
| E7 | Commits in a session's clone use `<agent>@<business>.mosaic.invalid`. Landed squashes use "Mosaic bus host" | `sandbox.mjs:151-157`, `working.mjs:280-281` |
| E8 | No code pushes to a remote. The git socket pushes bundles into a local bare working repo with no remote and no credential | `working.mjs:1-14, 198-238` |
| E9 | Adding or changing an agent refuses a definition that declares credentials (`organization-credentials-out-of-scope`). Binding them is a hand edit. Row 194's design keeps that refusal for create and hire "until the credentials row" | `business/src/instance-settings.mjs:20-29`, row 194 DESIGN.md D1 |
| E10 | The development repository's helper pushes as one identity from one JSON file | `scripts/git-credential-mosaic.mjs` |
So per-agent references and a host-only reader already exist for two kinds. What's missing: a fixed place for
the files, an agent identity, any remote push, lifecycle verbs, and a model credential that agents don't
share.
## 2. The store (brief item 1)
D1. **Layout.** Each stack instance keeps its agents' secrets under its own config dir:
```
<configDir>/secrets/agents/ 0700, owner = the stack instance's OS user
<configDir>/secrets/agents/<business>/<agent>/ 0700
<configDir>/secrets/agents/<business>/<agent>/<kind> 0600, the token only, one line
```
This is row 190's secrets root (`<configDir>/secrets/...`, as Q175's slot uses). The token file keeps E2's
format, so the bus reader doesn't change. Nothing in a token file is a name, an email or a URL. Those are not
secret and go in the business file (D6), where review sees them.
D2. **The business file reference.** A new reference form, `{"store": true, "rotateBy" | "expires": DATE}`,
means "the file at the layout path for this business, agent and kind". The path is derived, never written,
so an agent's entry can't name another agent's file. The old `{file}` and `{env}` forms stay valid for
migration (section 7). A `{file}` reference that points inside `<configDir>/secrets/agents/` but not at its
own agent's path refuses `credential-not-yours` when the business file loads.
D3. **Read path.** Unchanged in shape: the bus reads every entry once at start (E2) and on a store change
(section 6), and holds the values in memory. No sandbox binds `<configDir>` (E4 already holds). No token goes
into a sandbox for `gitea` or `vikunja`. The agent asks the bus to act, and the bus uses the token.
D3a. **A `store` reference with no file.** ENOENT on a `store` path loads as an absent entry. `status` shows
`not-issued`, and `useFor` refuses `credential-unavailable`. The bus still starts. This is the state after a
create without `issue` (D13) and after `revoke` (section 6). Every other defect still refuses start: wrong
owner or mode on the file or a dir, a symlink, a non-regular file, bad format, a missing or wrong-mode
`secrets/agents/` dir. `{file}` and `{env}` keep E2's behavior, where a missing file refuses start.
## 3. The "RLS equivalent" (brief item 1)
Three layers. Each one is enforced by code or the OS, not by prose.
D4. **OS, between stack instances.** Files are 0600 and dirs 0700, owned by the stack instance's own OS user.
Case 1's user can't read Case 2's store (Q180: each instance under its own user without sudo). The bus refuses
to start when any file or dir under `secrets/agents/` has the wrong owner or mode (`credential-file`, as E2
does for files today, extended to the dirs).
D5. **Bus, between agents of one stack instance.** The principal for every credential use comes only from the
verified cap (E5). The agent-facing API is `credentials.useFor(session, kind, op)`: `session` is the broker's
frozen record, and the key is built from `session.business` and `session.role`. It has no agent argument, so
there is nothing in a request to point at another agent. The existing `use(name, kind, op)` stays for the
host's own `@sync` entry and is not reachable from a request. The tasks adapter's per-role calls (`ctx.as`,
`verbs.mjs:126`) move to `useFor`. Any request field that names an agent other than the caller (`as`,
`agent`, `instance`, `role` on a credential path) is refused `credential-not-yours` and recorded through
`broker.refuse()` as `action.refused`, with the caller's cap and the task ref. A build test greps every
`credentials.use(` call site and fails on a new one outside the allowlist.
D6. **What stays open.** A process of the same OS user outside the sandbox can read every agent's file. That
is the same limit as the human proof (P19 333). Q180's per-user VMs move the boundary to "one stack instance
per OS user". Agents of one instance are kept apart by the sandbox and D5, not by the OS.
## 4. Kinds (brief item 2)
| Kind | 10-27 | Use |
|---|---|---|
| `gitea` | yes | Git push to a granted remote (section 5). Gitea API verbs (issues, pull requests) wait. The factory's review flow runs in the working repo today, and no 10-27 case needs an agent to open a pull request |
| `vikunja` | yes, exists | Moves to `store` references and `useFor`. No behavior change |
| `model` | per stack instance only | See D7 |
D7. **Model provider.** A model token has to sit inside the sandbox today, because the harness calls the
provider itself (E6). Per-agent model tokens need a proxy that holds the real token and gives each run its own
short-lived one, which is the CLIProxyAPI line in TARGET.md. That waits for its own row. For 10-27 the model
credential is per stack instance: row 190's `<configDir>/secrets/claude-session.env`, owned by that
instance's OS user. It is never a copy of jwoltje's file. Jason signs in on each VM user himself (A6). Whether
Case 1 and Case 2 run on Jason's existing subscription or need a new account is Jason's call (Q-J1). This row
doesn't pick. A pi model credential also waits for P19 #383.
## 5. Git identity and push (brief item 3)
D8. **Identity.** A new business-file key, `roles.<agent>.identity = {name, email, gitea: {url, user}}`. It is
optional, because today's business files with a `gitea` agent (coder) must load unchanged. `publish` (D9
step 2) requires it, and requires that the definition declare `gitea`. A business file with a `publish`
grant on an agent that lacks either refuses at load, and the error names `roles.<agent>.publish` and says
"needs identity and a gitea definition" (N10). Load is the only place this is checked: a business file that
fails it never reaches the bus. The sandbox sets `GIT_AUTHOR_*` and `GIT_COMMITTER_*` from
it in place of E7's invalid domain. The email is the one Jason gives the agent's Gitea user, so Gitea links the
commits to that user. Agents without `gitea` keep E7.
D9. **Push to a remote: a new git socket op, `publish`.** No new socket, no bundle, and no token in the sandbox.
```
publish {"cap", "op": "publish", "task_ref", "project", "branch", "decision"}
-> {"ok": true, "result": {"fork", "ref", "commit", "previous"}} or a refusal
```
The host checks, in order, and each refusal goes through `broker.refuse()`:
1. Identity from the cap (E5), then the broker's `#checkAuthority(s, "git.push.remote", {decision, target:
task_ref})`, as `checkLand` does for a merge (`broker.mjs:907-913`). `git.push.remote` is a new action in
ACTIONS and is not in ROUTED (REQ-WAIT-3, `packages/business/src/vocabulary.mjs:52-58`,
`broker.mjs:42-69`), so it is a hard wait and classifies `gated` for every role. Every publish therefore
needs a decision the human resolved for this task ref. A decision is spent once (`decision-consumed`).
**The decision binds one commit.** `#authorityFor` binds action, role, thread and target only
(`broker.mjs:769-798`), so the commit needs its own carrier. The raise records it the same way it records
`holderRun` for `role.revoke` (`broker.mjs:1733`). When the action is `git.push.remote`, the
`decision.raise` event body gets `pushCommit`. That is the commit of `#latestPush(business, target)` at
raise time, read by the broker and never taken from the agent. A raise with no recorded push for the task
refuses `no-push`. The decision's question shown to the human ends with that commit. `#authorityFor` then
refuses `decision-mismatch` when `context.pushCommit` differs from the commit of `#latestPush` now, which
is the same check `holderRun` gets at `broker.mjs:796`. So the sequence approve at A, push B, publish B
refuses.
No role definition grants the action yet. When Jason's PRD version routes it (Sage, Vikunja #420), a
reviewed commit adds it to both ROUTED lists and to coder's `withinRole`. From then on a coder publishes
without a decision, and `decision` is optional for a within-role class only.
2. The project is in the agent's grant, `roles.<agent>.publish: [project]`, and the business file names the
project's upstream repository, `projects.<project>.remote: "https://HOST/OWNER/REPO"`
(`remote-not-granted`). Both URLs pass D9c's parse at load. The upstream's host and port equal
`identity.gitea.url`'s (`remote-host-mismatch`). The host never contacts that URL. It builds the agent's
fork, `fork = https://HOST[:PORT]/<gitea.user>/REPO`, from the upstream's origin and repository name and
the agent's own `gitea.user`. `fork` is the only URL the rest of `publish` uses: D9a's `ls-remote`, step
5's push, the helper's host match, `checkPublish`'s result and step 6's event. So the owner of the
repository that receives a push is the agent's own Gitea user on every push, by construction, and the
agent can't name it (webby-0011 N1). D9c is what makes "by construction" hold. Gitea's refusal of a
push to a repository the user doesn't own or collaborate on is the second layer. An upstream whose owner
equals the agent's `gitea.user`, compared without case, refuses `remote-not-fork`, at load and at push. `remote` is a new key
beside `root` and `repo` (`business.mjs:186`), and `identity` and `publish` are new keys on a role entry
(`business.mjs:89`). All three are optional, so today's business files load unchanged.
3. The source is one commit, `C = context.pushCommit`. It is checked three ways, so all three are equal:
the decision's commit, the commit of `#latestPush`, and the work ref's tip in the working repo,
`work/vikunja-<project>-<task>` (`working.mjs:110-114`). The tip is read under the repository's lock, and
if they differ the push refuses `ref-moved`. The work ref is per task, not per agent. Two checks make it the
caller's: `#latestPush` was recorded by the caller's role (`work-not-yours`), and step 4. The destination is always `refs/heads/<gitea.user>/<branch>`, built by the host.
`validBranch(branch)` holds (`invalid-request`). The agent never names the destination ref.
4. Every commit in `main..<work ref>` has author and committer email equal to `identity.email`
(`identity-mismatch`).
5. The push runs through `credentials.useFor(session, "gitea", ...)`: `git push <fork> <C>:<dst>`, by oid,
with no `+` and no `--force`. It never pushes by ref name, so a push to the working repo after
`checkPublish` can't change what goes out (`non-fast-forward` from git stays a refusal). Git runs with E8's SAFE flags
(`working.mjs:92-98`), which set `protocol.allow=never`. `publish` adds exactly two flags,
`protocol.https.allow=always` and `http.followRedirects=false`, plus `GIT_TERMINAL_PROMPT=0` and no global
or system config. The token reaches git through a one-shot credential helper that reads it from an
inherited pipe, never through argv, env, the URL or a file. Seat `gitEnv` (`checkout.mjs:79-89`) takes only
PATH from the host and sets the rest itself, so no proxy or `GIT_SSL_*` variable reaches git. `publish` takes a `tls` parameter that
defaults to `[]`, the same way `push` takes `fsck = FSCK` (`working.mjs:100`). Tests pass
`["-c", "http.sslCAInfo=<scratch CA>"]`. The host never sets it, and nothing in config or env reaches it. The helper answers `get` only when `protocol`
is `https` and `host`, port included, equals `fork`'s. It writes nothing for any other `get` and for
`store` and `erase`. `assertClean` checks git's output, so a fork that echoes the token gets
`credential-leak`.
6. The bus records `action.allowed` for `git.push.remote` with fork, ref and commit.
D9a. **The agent's fork must share root's history.** A push sends every object the fork lacks. The working
repo's `main` is the project root's history (`working.mjs`, seed), so a fork that doesn't share it would
receive all of root, landing squashes included. The check runs on `fork`, never on the upstream: a related
upstream says nothing about what the agent's `<user>/REPO` holds.
`publish` runs `seed` first (`working.mjs:164-176`), under the repository's lock as `push` does, so `main`
catches up with root before the check below. That means a fork also receives any landed root commits that
upstream doesn't have yet, because they sit under the agent's work. That is expected: the agent's branch is
built on them.
Before the push, the host runs `git ls-remote --symref <fork> HEAD` (same flags and helper). It refuses
`remote-unrelated` when the fork is empty or when that tip isn't a commit the working repo already has and an
ancestor of `main`. When `ls-remote` exits non-zero, for a fork that doesn't exist, a refused token or a
network failure, `publish` refuses `fork-unavailable` and the push doesn't run (N11). Git's stderr isn't
parsed to tell those apart and isn't returned, since it can carry the URL. Only `ls-remote` crosses the
network for this, so nothing is fetched from the fork. If the fork has moved past root, the refusal stands
until root catches up. That is a human pull into root, not an agent's.
D9c. **The URL parse (B6).** `projects.<p>.remote` and `identity.gitea.url` are parsed with `new URL()` at load,
and the business file refuses to load unless all of these hold:
- The scheme is `https:`. There is no username, password, query or fragment.
- The string equals its own normalized form: for the upstream, `url.origin + url.pathname`; for
`gitea.url`, `url.origin + "/"` or `url.origin`. So a dot segment (`/<user>/../OWNER/REPO`), a
percent-encoded dot, an upper-case host, an explicit `:443`, a backslash, userinfo, a query and a fragment
each refuse, because `new URL()` would change them (checked with node: each differs from its normalized
form). A trailing slash survives normalization and is refused by the two-segment rule below. The checks below run on that path, so the owner that is checked is
the owner that is used.
- The upstream's path is exactly `/OWNER/REPO`: two non-empty segments, no more and no fewer. One trailing `.git` on
REPO is allowed and stripped. `fork` is built from the stripped REPO.
- `OWNER`, `REPO` and `gitea.user` each match Gitea's name rule, `^[A-Za-z0-9_.-]{1,100}$` and not `.` or
`..`. REPO after stripping must still match.
- `gitea.url`'s path is `/`. A Gitea served under a sub-path isn't supported by this row and refuses at load.
- Host and port are compared after `new URL()` normalized them, and an omitted port counts as 443 on both
sides.
- Gitea compares user and repository names without case (`lower_name`), so `remote-not-fork` compares the
upstream's OWNER with `gitea.user` without case. An upstream owned by `Coder` and a `gitea.user` of `coder`
would otherwise build a "fork" that is the upstream itself.
D9b. **How `publish` reaches the broker and the store.** The git socket runs in the bus process
(`cli.mjs:504-506`) and reaches the broker only through the host's op table (`bus/src/runtime.mjs:222-246`).
`publish` adds three entries there, each wrapped in `refusing()`:
- `checkPublish(cap, args)`: steps 1 to 3, including the decision check. It returns `{ticket, fork, user,
email, source}` and no token. The ticket holds the session and the decision until `recordPublish`, the
same way `checkLand` holds a landing.
- `withGitea(cap, fn)`: `credentials.useFor(broker.session(cap), "gitea", fn)`. `fn` gets the token and runs
step 5. The token never leaves that call.
- `recordPublish(ticket, result)`: step 6, and it spends the decision.
The op table gets no entry that returns a token.
D10. **Which refs.** Our side: only `refs/heads/<gitea.user>/*`, set by the host (D9 step 3). The default
branch and other agents' prefixes can't be named. Gitea side, two options (Q-M1):
- **O1. A fork per agent. Ruled by Mos (mos-1027).** The agent's Gitea user has read on the upstream repo and owns a fork
named like it. The host derives the fork's URL from the upstream and the agent's user (D9 step 2).
Its token can write only repos the user owns or collaborates on, so Gitea itself stops any push to upstream
or to another agent's fork. No branch rules are needed per agent, so none can drift. The cost: work reaches
upstream as a pull request from the fork, opened by a human or a later API verb.
- **O2. One shared repo. Not chosen.** The agent is a Write collaborator. Upstream's default and release branches get
branch protection with no agent on the push list, plus a rule `<user>/*` per agent. Gitea doesn't stop agent
A from creating an unprotected name outside `A/*` unless a catch-all rule outranks the per-agent ones. That
depends on Gitea's rule priority, which I haven't verified on our version (probe P1). Every hire adds a rule.
Token scopes for either: `write:repository` and `read:user`, nothing else, per vocabulary.mjs rules (no
`all`, no admin).
D11. **The host git socket.** `publish` is added to the socket. `push`, `fetch` and `land` don't change. The
socket's bind rule (E4) adds `git.push.remote`. The working repo stays credential-free and remote-free. The
`fork` URL is passed per call and never written into its config. A landing still squashes into the project
root as "Mosaic bus host". Landings don't push anywhere.
## 6. Lifecycle (brief item 4)
All verbs are on the bus, in the CLI (`mosaic credentials ...`) and in the Console's agent page. Every verb is
human-only. Under Q173 the tester may run `status` and `revoke` on the dev stack instance. `issue` and
`rotate` stay with Jason, because the tester is an agent and the token would pass through its session
(duckworth N1). Each one appends to the organization
writer's `changes.jsonl` (actor, agent, kind, verb, date; never a value). No schema migration, no new event
kind. A refusal uses `action.refused`.
| Verb | What it does |
|---|---|
| `issue AGENT KIND --rotate-by DATE` (gitea) or `--expires DATE` (vikunja) | Reads the token on stdin (never argv), checks E2's format, writes `<kind>.tmp` 0600, fsyncs, renames over `<kind>`, and sets the business file reference to `{"store": true, date}`. The bus reloads that one entry, with no restart |
| `rotate` | Same as `issue` over an existing entry. The old value is gone from disk and memory when the call returns. Jason then deletes the old token in Gitea (A5) |
| `revoke AGENT [KIND] --reason` | Drops the entry from the bus's memory first, then unlinks the file. A use in flight finishes. Every later use refuses `credential-unavailable` |
| `delete AGENT` | `revoke` for every kind, then removes the agent's dir. Refused while the agent has a running claim |
| `status [AGENT]` | Kind, date and state per entry (E3's `status()`), no values |
D12. **"Stops working at once."** On our side this holds without a timing window. A fired agent never holds a
`gitea` or `vikunja` token (D3), so revoking means the bus no longer has it. On the Gitea and Vikunja side the
token still exists until Jason deletes it (A5). The status line for a revoked entry says so until Jason
confirms. The model token is the exception: it sits in the running sandbox (D7), so a fired agent's runs must
end. Row 194's fire ends them.
D13. **What row 194 calls.**
- `credentials.issue` is never called by the chief. Row 194's hire stays uncredentialed (its D1).
- A human `create` of a credentialed definition: row 194 lifts E9's refusal for the human path only. The agent
is created with `store` references and no files. `launch` refuses `credential-unavailable` until Jason runs
`issue`.
- `fire` and `delete`: row 194 calls `credentials.revoke(business, agent, {reason, actor})` before it changes
the business file. A failed revoke stops the fire (`credential-revoke-failed`), so an agent can't be fired
with its credentials still live on our side.
- `regenerate`: no credential change.
## 7. Who does what (brief item 5)
Agents never mint accounts or tokens. Jason's steps, per agent that needs `gitea`:
- A1. Create the Gitea user: Site Administration, User Accounts, Create User Account. The name is the agent's
`gitea.user`. The email is the agent's `identity.email`.
- A2. Create its token: sign in as that user, Settings, Applications, Generate New Token, scopes
`write:repository` and `read:user`, nothing else.
- A3. Create the fork of each granted repo as that user (O1). Gitea's push-to-create
(`ENABLE_PUSH_CREATE_USER`, `ENABLE_PUSH_CREATE_ORG`) stays off, so a push to a missing `<user>/REPO`
fails rather than creating a plain repository (N11). `publish` refuses `fork-unavailable` before that.
- A4. On the stack instance's VM, as its OS user: `mosaic credentials issue AGENT gitea --rotate-by DATE`, with
the token pasted into stdin.
- A5. On fire or rotate: delete the token in Gitea (Settings, Applications), or disable the user.
- A6. Model: sign in to the provider on each VM user to write `<configDir>/secrets/claude-session.env`.
Vikunja bot accounts and tokens are already Jason's (tracker bots). Only A4 changes for them.
## 7a. The grant path (mos-1124, Q188, Q190)
Q188: no agent creates a grant. The chief raises the request, Jason approves it in the inbox, and protected
stack code creates the grant. Jason called manual grant steps a no-fly, so A1 to A4 above are the fallback only.
- a. `credentials.issue`, and so `rotate`, accepts a resolved `credential.mint` decision, Jason's Approve, as
the human path. #436b's executor calls it with that decision and nothing pasted. #436b adds that path and
the decision kind. Row 199 ships the verbs with today's human check only.
- b. Under Q190 ("Q12: O1"), each agent seat gets its own Gitea account. On the Approve, #436b's executor does
A1 to A3: the seat's Gitea user, its token and its forks. Until #436b lands, pushes stay as jason.woltje.
- c. Q185 settles Q-J1 and Q190 settles Q-J2.
No code in 07141966e changes for this section. The executor and the decision route are #436b's.
## 8. Migration (brief item 6)
- M1. Existing `{file}` and `{env}` references keep working. A business moves one agent at a time with
`issue`, which rewrites that agent's reference to `store`.
- M2. Nothing here touches the dev repository's helper, its `.git/config`, `~/.mosaic` or `~/.mosaic-dev`.
Dev seats (jarvis, seat tokens) keep pushing as today. The mos-0981 rule ends only when Mos says so.
- M3. The live dev stack instance needs no change at re-pin: no reference uses `store` until someone issues
one.
## 9. Questions
Mos ruled on the authority model in mos-1027 (2026-10-20T19:27Z):
- Q-M1. O1, one fork per agent.
- Q-M2. `git.push.remote` and its bounds in D9 to D11 are approved. It lands as a hard wait: out of ROUTED,
and no role grants it within-role (D9 step 1). Sage puts its routing in the PRD version draft going to
Jason with `role.hire` and `role.fire` (Vikunja #420). After his approval, a reviewed commit routes it and
grants it to coder.
- Q-M3. Issue, rotate and delete are human-only, and the chief gets none of them. Under Q173 the dev
instance's tester gets `status` and `revoke` only.
- Q-M4. The model credential stays per stack instance for 10-27. Per-agent model tokens wait for a proxy
row.
Open for Jason. Mos takes these to him one at a time:
- Q-J1. The model account for Case 1 and Case 2: existing subscription or a new account. Settled by Q185 (7a).
- Q-J2. Gitea users and tokens for the agents that need them (A1 to A3). Settled by Q190 (7a).
## 10. What this closes (brief item 7)
- G11: closes the store (D1 to D6) and lifecycle (section 6) for `gitea` and `vikunja`. Still open: per-agent
model credentials (D7), and the same-uid limit inside one stack instance (D6).
- G13: closes "own credentials for git" (D8 to D10). Still open: Gitea API verbs (section 4).
- C2.5 (git remote): Case 2's books repo can use `publish` once its project names a remote.
## 11. Tests and mutants
All tests use scratch roots, fake tokens and a fake Gitea: `git http-backend` behind a node http server on
127.0.0.1 that checks basic auth against a fake token and records each ref update. No request leaves the host.
- T1. Layout: wrong mode or owner on a dir or file refuses start. A `{file}` reference into another agent's
path refuses `credential-not-yours`. A bus restart after a create without `issue` starts, and the agent's
status is `not-issued`. A missing `{file}` still refuses start.
- T2. `useFor` keys from the cap only. A request naming another agent is refused and leaves one
`action.refused`.
- T3. The fake Gitea serves https on 127.0.0.1 with a cert from a scratch CA made by the test. It reaches git
only through the `tls` parameter (D9 step 5). Each refusal test that should reach the network asserts that
the first port received the request, so no test passes because git refused http first. `publish`: the
happy path lands `refs/heads/<user>/<branch>` on the agent's fork in the fake, which shares root's history, at
exactly the decision's commit. Refusals for each step in D9, one test each, including `work-not-yours`. Until routing, a publish with
no decision refuses, one with a decision for another task or commit refuses, and a second publish on a
spent decision refuses `decision-consumed`. Approve while the latest push is A, push B, publish: refuses
`decision-mismatch`, and the fork receives nothing. A push to the working repo between `checkPublish` and
the git push refuses with `ref-moved` or `decision-mismatch` (sage-1440, BUILD.md deviation 14). A raise with no recorded push refuses `no-push`. A test pins that `git.push.remote` is in neither ROUTED list
and that no `roles/*.json` grants it. A force-push attempt is
refused by git's non-fast-forward. A fake fork that answers 302 to a second 127.0.0.1 port: the push
fails, and the second port receives no request with an Authorization header. A remote on the right host
and a different port is refused at step 2. Two agents granted one project each land on their own fork:
agent A's push reaches `/<A's user>/REPO` only and B's reaches `/<B's user>/REPO` only, and the fake's
upstream repository receives nothing. An upstream owned by the agent's own user refuses `remote-not-fork`,
and so does one whose owner differs from `gitea.user` only in case. With an upstream that shares root's
history, a first publish to an empty fork and to an unrelated fork each refuse `remote-unrelated`, and
neither fork nor upstream receives objects. A fork that doesn't exist refuses `fork-unavailable`, and the
fake records no push. `ls-remote` and the push each reach the fork's path only (the fake records every
path). Load refusals (D9c), one test each: a dot segment, an extra path segment, a missing segment,
a trailing slash, userinfo, a query, a fragment, an explicit `:443`, a name outside Gitea's rule, and a `gitea.url` with a
sub-path. A `.git` suffix loads and builds the fork without it. A `publish` grant on an agent without
`identity`, and on one whose definition lacks `gitea`, each refuse at load with the N10 message.
- T4. Token handoff: the token never appears in argv, env, the fork URL, the working repo config,
`changes.jsonl`, events or logs (scan every file under the scratch roots).
- T5. Lifecycle: issue, rotate, revoke and delete, each live with no restart. After revoke, `publish` and the
Vikunja path refuse `credential-unavailable`, and so they do after a bus restart that follows the revoke.
The tester can run `status` and `revoke`, and `issue` and `rotate` refuse it.
- T6. Default: an agent without `identity` commits as E7. Existing `{file}` references load unchanged.
Mutants to kill: drop the cap-derived key in `useFor`, let the request name the destination ref, add `+` to
the refspec, skip the identity check, put the token in the URL, revoke file before memory, skip the dir mode
check, allow a `{file}` reference into another agent's dir, make ENOENT on a `store` path fatal again, make
ENOENT on a `{file}` path silent, a helper that answers any host, drop `http.followRedirects=false`, drop the
`work-not-yours` check, drop the `remote-unrelated` check, add `git.push.remote` to ROUTED, skip the spent
check on the decision, drop the `pushCommit` comparison in `#authorityFor`, push by ref name in place of the
oid, push to the upstream URL in place of the derived fork, run `ls-remote` against the upstream in place of
the fork, take the fork's owner from anywhere but the agent's `gitea.user`, compare the owner with case, run the
D9c checks on the raw path in place of the normalized one, read a failed `ls-remote` as empty, skip the tip
check under the lock, and pass a CA flag in production.
## 12. Risks
- R1. Backups: if a backup tool copies `<configDir>`, it copies tokens. The build checks row 185's backup set
and says what it includes.
- R2. Withdrawn. O1 was ruled, so probe P1 isn't needed.
- R3. Outbound HTTPS from the bus host to Gitea is new network reach for the host process. On the VMs this is
Jason's network to confirm.
+10
View File
@@ -0,0 +1,10 @@
# Quackerjack
Builder seat, Claude Haiku 5.5 at high effort, started by Sage under mos-1023 item 1 and Q177 for federation
rows R5 (`mosaic access list, grant, revoke`) and R2 (the `federation` config key), per Gizmoduck's design
section 7 (#414).
No launcher in this directory. The seat runs as a T3 thread and works in its own worktree. It has no
credentials of its own; until the credentials row settles per-agent identity, it pushes only its own
`quackerjack/*` branches through the repository credential helper (mos-0981), never `refactor`, `main` or
another seat's branch.
-6
View File
@@ -238,12 +238,6 @@ Exit codes: `0` ok · `2` invalid file or credential reference · `3` system
config problem · `4` usage or a required file missing. Details:
`packages/business/README.md`.
`mosaic credentials status|issue|rotate|revoke|delete` (row 199) goes to
`packages/cli` and the bus's human-only credential verbs: an agent's own
token in `<configDir>/secrets/agents/<business>/<agent>/<kind>`, read on
stdin, never printed. The tester may run `status` and `revoke`. Details:
`packages/cli/README.md`.
`instances` and `instance add|set` (row 165) go to `packages/cli` and the
bus's human-only `instances` and `instances.set` verbs, so they need a live
bus host and refuse inside an agent session. `instances` prints the revision,
@@ -0,0 +1,120 @@
# Federation build rows R1, R2 and R5 (#414)
Sage, 2026-10-20, for Mos's orders mos-1023 item 1, mos-1028 item 2 and mos-1034 item 1. The design is
Gizmoduck's `agents/gizmoduck/work/federation/DESIGN.md` at a1daa3c07 on `gizmoduck/federation-design` (r4.9,
sha256 a31cbebd070f9e3a5ca4ccb158b1747c6f5a83c3469855705b0026a581fb5c6a). Duckworth approved r4.5
(duckworth-0012), r4.6, r4.7 (duckworth-0020) and r4.8 (duckworth-0021, 8a22d0897 on `duckworth/reviews`),
and acked r4.9 (duckworth-0024). r4.7 applies Jason's Q181 hosts, r4.8 adds N22's R8 checks and r4.9 names who
runs them. None of them changes R1, R2 or R5. Section 7's table
sets each row's scope and test gate. Vikunja P19 #414. Seats push only their own `<seat>/*` branches
(mos-0981).
The other R rows are filed when their parents have candidates: R3 after R1 and row 196, R4 after R3, R5 and
row 193's viewer level (193a, mos-1019), R6 after #412 phase 1, R1, R2 and R3, R7 after R1-R6 and row 190.
R8 is a live window, not a code row. Its brief is the last section here.
## Brief: FED R1, the access.json loader and validator
Owner Gizmoduck, reviewer Duckworth, gate owner Sage. Base: refactor at this brief's commit. No parent.
`packages/business/src/access.mjs`: the access.json loader and validator from DESIGN.md section 3.5, and the
pure `covers`. Providers, peers and businesses come in as arguments. One read plus the file checks. The
exported interface is what R5 and row 193a build against, so the manifest names it, and any change to it in
review goes to Sage before it lands. Section 7 names `covers` but not the loader and validator exports or
the context shape. Morgana's 193a (508ee4d61 on `morgana/row193a`) assumes `checkAccess(doc, {providers,
peers, businesses})`, with `businesses` mapping an id to `{human, instances}`, throwing a `BusinessError` that
names the key path. Use those names, or tell Morgana and Quackerjack yours before the manifest goes to
Duckworth. Row 193a later widens the level check to `viewer` on R1's candidate (mos-1019).
Tests, in `packages/business/tests/`: every rule in 3.5 refuses with the key path named. That covers a
symlink, a non-regular file, another owner, group or other write, expiry, `*`, a grant disjoint from the given
subjects, `human` and `tester` ids, any level other than `collaborator`, and a project entry that isn't a
UUID. For `covers`: `*` and listed ids match, while an unlisted project, instance or business doesn't, an
expired grant doesn't, and a level filter rejects another level. One mutant per refusal, each killed.
Boundaries: no change to `~/.mosaic-dev`, `~/.config/mosaic-dev`, the live units or any live access.json. Tests
use scratch roots. Single test files run freely, and suites need Sage's slot.
Gate: Duckworth approves the manifest. The business suite and every `scripts/test-*.sh` are green on Sage's
gate, and the mutants are killed.
## Brief: FED R5, mosaic access list, grant and revoke
Owner Quackerjack, reviewer Duckworth, gate owner Sage. Chained on R1 (Q177). The build starts now against
R1's exported interface as the R1 manifest names it, and R5 lands after R1 and rebases onto R1's landing.
`mosaic access list`, `grant` and `revoke`: human proof, atomic write, revision bump, a receipt line. The help
text states the `instances` limit and the provider-removal order (DESIGN.md section 7, row R5). R4 later makes
these verbs call `access.reload`, so R5 leaves no other reload path.
Context (sage-1419). `ssoSubjects` is row 192's `console.sso.providers[].subjects` flattened to `{provider,
sub}` pairs; the design's "`console.sso.subjects`" in 3.5 is shorthand for it. `providers` is the list of
`providers[].id`, read through `loadSystem` and row 192's `checkConsole`. `peers` is `[]` until R2 lands, so a grant
naming a peer home refuses. R5 rebases onto row 192's landing as well as R1's.
Tests, in `packages/cli/tests/`: tester and agent callers are refused, a concurrent write refuses on a
revision conflict, the receipt line is appended, and a bad grant is refused before anything is written. One
mutant per refusal, each killed.
Boundaries: no change to `~/.mosaic-dev`, `~/.config/mosaic-dev`, the live units or any live access.json. The
human proof in tests uses the scratch stand-in pattern (a scratch bus verifies its own human-cli path).
Single test files run freely, and suites need Sage's slot.
Gate: Duckworth approves the manifest. The cli suite and every `scripts/test-*.sh` are green on Sage's gate,
and the mutants are killed.
## Brief: FED R2, the federation key in the system config
Owner Quackerjack, reviewer Webby, gate owner Sage. Chained on row 192 (#412 phase 1). The build starts when
row 192 has a candidate, on that candidate.
A `federation` key in `scripts/mosaic-config.mjs` with the shape in DESIGN.md section 3. A config with no
`federation` key still loads as it does today. This changes the config schema, so the manifest lists every
reader of the key.
Tests, in `scripts/test-config.sh`: a valid key loads, an unknown sub-key and a bad origin refuse, and a
config without the key loads. One mutant per refusal, each killed.
Boundaries: `~/.config/mosaic-dev/config.json` is user-authored and is not touched. Tests use scratch config
dirs only.
Gate: Webby approves the manifest. The cli suite and every `scripts/test-*.sh` are green on Sage's gate, and
the mutants are killed.
## Brief: FED R8, the hub window on 10.1.1.54
A live window, not a code row (DESIGN.md section 7, row R8). The brief follows r4.11, b93f098f0 on
`gizmoduck/federation-design` (sha256 1bcee8df0a3a8628a6646535b756c936ca6d9e7bcfa7e1dd2bf1d3f01715c8e6),
which Duckworth acked with r4.10 (duckworth-0034). Mos closed design rounds there (mos-1072). Run by Jason and
Mos. Nothing here authorizes a step on any host; each step waits for the window's go.
What changed since r4.9 (DECISIONS Q182, Q183, mos-1055, mos-1063). R8 is a recorded acceptance. It starts
with Jason's acceptance of the factory paths to .54 on record, quoted in section 8. The privilege and host
checks still run, and each result goes into the R8 record with the name of whoever ran it. They report and
don't refuse. The four functional checks named at the end of section 7's R8 row still stop R8 when they fail
(mos-1066 item 1).
Order (mos-1042 item 2). R8 runs after the Case 2 install on .54 (#416, due by 2026-10-27, Jason's latest-by
date in Q176, mos-1069), after Bean and Chet are live in Case 2, and after both Hermes gateways on 10.1.1.54
have stopped. Case 2 go-live doesn't wait on R8. The risk, stated for 10-27: federation between the two
instances depends on Melanie's switch-over date.
Entry checks, all before anything starts:
- R7, #412 phase 2 and Jason's #419 answer, as section 7 lists.
- The four hub user checks on 10.1.1.54 (N22, mos-1035 item 1): `sudo -l -U mosaic` grants nothing, `id
mosaic` shows only its own group, `~mosaic/.ssh/authorized_keys` is absent or empty, and no file under
`/etc/sudoers.d` names `mosaic`. An agent or Jason runs them as localadmin (mos-1055), and the output goes
into the R8 record with who ran them.
- The factory privilege checks, the key fingerprint comparison and the section 8 privilege path table, each
as section 7's R8 row states them.
Build notes:
- N1 (Duckworth, duckworth-0034, mos-1072). A check that fails on a path Jason didn't name in Q182 or Q183 is
marked "not named in Q182 or Q183" in the record, not filed under the acceptance. Examples: a factory mount
of hub data, another agent on .54, a provider admin token. Mos takes each one to Jason.
- N2 (Duckworth, doc-only). The Q183 and Q4 quotes in section 8 are excerpts. When the R8 record quotes them,
it marks each omission with "…" and checks the full text at DECISIONS.md:1535-1541 in jason-master-restart
(read only).
Done when every entry check has its output recorded, Jason signs in to the hub and sends one message, Melanie
revokes, and his next request is refused.
+81
View File
@@ -0,0 +1,81 @@
# Brief: a pre-push hook that refuses force-pushes, and a busy check that fails on BUSY (#426)
Sage, 2026-10-20, for Mos's order mos-1078 (Vikunja P19 #426) and its go mos-1083 item 1. Owner Rocko,
reviewer Dewey, gate owner Sage. Base: refactor at this brief's commit. Starts after row 191.
## Why
Morgana force-pushed a seat branch on 2026-10-20 (morgana-0016). Mos took it to Jason. Force-push is a
Q173 exclusion, and canon says never force-push, but nothing mechanical stops one. Mos: "Done means a
force-push from a seat worktree is refused, with a test that shows it."
Earlier the same day, a busy check chained as `| tail -1 &&` passed while the slot was BUSY. The script it
called prints BUSY or CLEAR, but it exits 0 either way.
The guard has two layers:
- (a) Gitea protection with force-push off on seat branch patterns and on refactor, next and main. ops-01
sets it up, through Mos (mos-1082). That part isn't this row.
- (b) The client hook below. It can be bypassed with `--no-verify`, so (a) is the real guard and (b) is the
early refusal.
## What ships
1. `scripts/git-hooks/pre-push` (new). For each ref line git passes on stdin:
- It refuses when the remote sha isn't all zeros and isn't an ancestor of the local sha (`git
merge-base --is-ancestor`). The refusal names the ref and both shas, and the hook exits 1.
- A new branch passes. A delete (local sha all zeros) is refused.
- Any failure of the check refuses too, for example an unknown remote object. The hook fetches nothing.
If the remote tip isn't present locally, it refuses and says to fetch first.
2. Install, following `scripts/queue-commit.sh --install-hook`. Either add a `--install-push-hook` there or
ship a small `scripts/install-git-hooks.sh`, whichever fits the existing code better. Say which in BUILD.md.
- The install copies the hook into the common `.git/hooks`, which every seat worktree shares.
- It refuses when `core.hooksPath` is set (the husky trap, 2026-10-09). It refuses when an installed
pre-push differs from the repository copy, unless `--force-install` is given.
- It never edits the installed copy in place.
3. `scripts/slot-busy.sh` (new), the shared busy check.
- It prints one BUSY line per matching process, or CLEAR.
- It exits 0 on CLEAR, 1 on BUSY and 2 on its own error.
- It matches `node --test`, `gate.sh`, `mutant`, `scripts/test-*.sh` and `docker compose` command lines.
It skips its own process and shell snapshots.
- It reads `/proc/<pid>/cmdline` and the cwd link only, never file contents under another seat's
directory.
- Usage in `docs/TOOLS.md`: `bash scripts/slot-busy.sh > LOG && gate`, with `set -o pipefail` in any
script that pipes it. Never `| tail -1 &&`.
## Tests
Add them to `scripts/test-land-work.sh` or a new `scripts/test-git-guard.sh` (say which). Run against a scratch
bare remote and a scratch clone with a worktree, never against origin:
- A rewritten branch pushed with `--force` from a worktree is refused, and the remote ref is unchanged.
- A `+refspec` push is refused.
- A fast-forward push passes.
- A new branch passes.
- A delete is refused.
- A remote tip that's missing locally is refused.
- The install refuses with `core.hooksPath` set, and refuses a differing installed hook.
- `slot-busy.sh` exits 1 with a fake `node --test` process running (a sleeping stand-in with that argv),
and 0 without one.
- `slot-busy.sh | tail -1 && false` doesn't apply here. The test runs `bash scripts/slot-busy.sh >/dev/null
&& echo ran`, and asserts it doesn't print `ran` while BUSY.
Mutants, each killed:
- drop the ancestor check
- treat an unknown remote object as passing
- let a delete pass
- install over a set `core.hooksPath`
- `slot-busy.sh` exits 0 on BUSY
## Boundaries
- Install the hook in the canonical checkout only after landing, as its own step with Sage. Tests use
scratch repositories.
- No push to origin from a test. No change to Gitea config. That's ops-01's.
- Don't touch `~/.mosaic`, `~/.mosaic-dev` or `~/.config/mosaic-dev`.
## Gate
Dewey approves the manifest. The new or changed `scripts/test-*.sh` file and every other `scripts/test-*.sh`
are green on Sage's gate, and the mutants are killed. Done when the hook is installed in the canonical
checkout and a scripted force-push from a seat worktree to a scratch remote is refused there. ops-01's
server proof (a refused force-push, plus a fast-forward push and a new branch that go through) is recorded
in Vikunja #426 by Mos.
+70
View File
@@ -0,0 +1,70 @@
# Brief: agent host access with sudo through the credential store (#424)
Sage, 2026-10-20, for Jason's rulings Q183 and Q184 relayed by Mos (mos-1056 item 3, mos-1067). Gitea #1707, Vikunja P19
#424, due 2026-10-27 (Q184: "Q5: by 10.27"). Owner Gizmoduck (design and build), reviewer Duckworth, gate owner
Sage. Chained on row 199's candidate (Q177): row 199 design r6 (4d3007ec9 on `megavolt/row199`, approved in
duckworth-0027) is not reopened. This row lands after 199 and rebases onto its landing.
## What Jason asked for
Q183, through Mos: "I envision a credential store within Mosaic Stack that works as some sort of variable for
the agent to call. The system will enforce access to the credentials to keep the agents sandboxed with proper
security... If we can utilize the credential store in Mosaic Stack and allow agents to perform their jobs with
NOPASSWD sudo on various machines, I am a happy man." Q183 also says not every agent needs that access.
Mos's requirements (mos-1067 item 1):
- An agent asks for a host credential by name.
- The stack checks a grant for that agent and that host, and does the ssh and sudo for it.
- The agent never sees the key or any secret value.
- Only agents with a grant can do it.
The PRD text is REQ-CRED-3, drafted by Honker for PRD 1.7 (sage-1422). If it misses 1.7 it goes into 1.8, and
this row lands only after the version that holds it is approved.
## What row 199 gives and what it doesn't (megavolt-0019)
It builds on 199 as 199 stands: the store layout under `<configDir>/secrets/agents/<business>/<agent>/`, the
lifecycle verbs and their `changes.jsonl` receipts, `useFor(session, service, fn)` with the D5 allowlist test,
`assertClean` on every result, and the grant shape of `publish` (a list on the role entry, checked by the
broker, plus a hard-wait action that isn't routed).
This row's own changes:
- A host kind in the closed kind set (vocabulary.mjs `SERVICES`, credentials.mjs, the broker's service check,
the CLI date flag, the remove regex), with a naming rule for one credential per host or per agent.
- A per-kind value check. An OpenSSH private key fails today's `TOKEN` pattern.
- Handing the key to ssh without the agent seeing it: an ssh-agent on a private socket, or a 0600 temp file
removed on every path, the way publish's one-shot helper works.
- Pinned host keys per host, and no trust on first use.
- The command boundary: what an agent may run under sudo, and how output returns without echoing the secret.
This is the main design work.
- The executor: a ticket and a host op running in the bus child, the same split row 199 uses for publish
(Megavolt's deviation in 199's BUILD.md).
## Steps
1. DESIGN.md first, in `agents/gizmoduck/work/queue-ROW/`. It answers each item above, names the grant shape
(`roles.<agent>.hosts` and a `host.sudo` action, or a reasoned alternative), says who grants (Jason, through
a reviewed role change, unless REQ-CRED-3 says otherwise), and lists tests and mutants. Duckworth approves it
before code. Sage reads it too, because it gives agents sudo on hosts.
2. Build on 199's candidate. Single test files run freely. Suites and mutant batches need Sage's slot.
3. The live part is not this row. Putting a key in the store and granting a host is a separate window with
Jason or Mos, after landing.
## Hosts
The first targets are `localadmin` with NOPASSWD sudo on 10.1.1.53 and 10.1.1.54 (#423 step 1, applied by
ops-01 at 20:05:47Z), and the servers agents already reach as localadmin (Q172). ops-01 owns the host side.
Tests use scratch sshd fixtures on 127.0.0.1 or fakes, never a real host.
## Boundaries
- No change to `~/.mosaic-dev`, `~/.config/mosaic-dev`, slot files, the live units or `~/.mosaic`. No key is
read, copied or printed. No ssh to any real host from a test or a seat.
- No widening of row 199's kinds or regex inside row 199. Both rows touch them, so this row rebases after 199
lands.
## Gate
Duckworth approves DESIGN.md, then the manifest. The bus, business and cli suites and every
`scripts/test-*.sh` are green on Sage's gate, and the mutants are killed (one per refusal, plus one that leaks
the key into the agent's output and one that skips the host key pin).
+49 -60
View File
@@ -1,77 +1,66 @@
# Brief: Melanie's hub instance on its own host (row 195, was 190b)
# Brief: the Case 1 install on 10.1.1.53 (row 195)
Sage, 2026-10-20. Rewritten for Mos's rulings mos-0998 and mos-1005 (Vikunja #418, #419). The first version
put the hub under its own Linux user on dragon-lin. That form can't pass Duckworth's B7 there, because the
factory uid's `sudo -n` succeeds on dragon-lin (Mos's recon, brain-redux
`projects/mosaic-mvp/MELANIE-HOST-RECON.md`). Melanie's own server is ruled out too: it has 1.2 GiB RAM free
with her two Hermes bots running, and co-tenancy would share her user and her passwordless sudo. The hub runs
on a separate host. Which host is Jason's decision (#419): hardware outside the cluster, a cluster VM with
factory PVE root removed, or a cluster VM with that residual accepted. This row is written so the host choice
doesn't change it. Gitea #1693. On the 10-27 MVP path. Federation R8 (Gizmoduck's design r4, 04a6112be) needs
this row or nothing.
Sage, 2026-10-20, third version, for Darkwing's rewrite request and sage-1439. Gitea #1693.
Owner Darkwing (Opus 5.5), reviewer Duckworth (author of B5 and B7), gate owner Sage. Chained on row 190
(Q177): the hub runs under row 190's config canon. Design first. The first deliverable is
`agents/darkwing/work/queue-190b/DESIGN.md`.
Why it changed:
- The first two versions put Melanie's hub on its own Linux user, then on its own host, Jason's choice under
#419.
- Q181 settled the hosts: Case 1 (Jason's software factory) runs on 10.1.1.53 and Case 2 (Melanie's hub) on
10.1.1.54.
- Q182 and Q183 made the hub's privilege checks a recorded acceptance. They report and don't refuse
(federation R8, mos-1055, mos-1063, mos-1066).
- The Case 2 install on .54 is Gosalyn's C2-13 plus its window (#416, due by 2026-10-27). It is not this row.
This row now installs Case 1 on .53. On the 10-27 MVP path (mos-0974, T1 item 1).
Owner Darkwing (Opus 5.5), reviewer Duckworth, gate owner Sage. Chained on row 190's candidate (the config
canon) and on C2-13's candidate (the install script), Q177. It reuses C2-13's `case2-install.sh`,
`case2-precheck.sh` and `case2-unit.sh` by input, not by copy. A gap in them goes to Gosalyn as a C2-13
change, or into this row's manifest as a named, reviewed change to those files. Design first:
`agents/darkwing/work/queue-195/DESIGN.md`.
## Goal
Melanie's hub instance (Case 2, C2-13) runs on its own host. No factory agent can read or write the hub's
config, data root or secrets, reach its bus socket, or log in to the hub host. Her writing reaches the hub only
through the existing git.mosaicstack.dev remotes, books repo only.
Jason's software factory runs on 10.1.1.53 as `mosaic` under `/srv/mosaic`, in the same unit form as Case 2
(design 9, Q181). It uses row 190's config canon and the model account the factory uses today (Q185,
mos-1081 item 1). No new provider account.
## Provision (blank until Jason answers #419)
## Provision
Who provisions the host and how is left open. Sizing from Mos (mos-0998): 4 vCPU, 8 GB RAM, 64 GB disk,
internal network only. Each agent process measures 300 to 390 MB RSS on dragon-lin. The design takes the host
as given: a fresh Linux host with a named admin path, reachable by the deploying seat while the row runs.
- ops-01 runs the root steps over ssh as `localadmin`: `useradd mosaic`, `/srv/mosaic` and the unit through
`case2-unit.sh` from a root-owned checkout. #423 step 1 (localadmin with NOPASSWD sudo on .53 and .54) was
applied at 20:05:47Z (mos-1066, mos-1077), and nothing else in #423 gates this row.
- #423 step 4 (jarvis sudo) waits until Case 1 is live on .53.
- The host list comes from Q181. Agents may run read-only checks on .53 as `localadmin`.
## The design must state
1. The deploy. Exactly what Darkwing installs on the hub host and how: the stack copy, node, the units, the
config dir and data root under row 190's canon (O1: `~/.config/mosaic-NAME`, and "`~` is the home of the OS
user that runs the instance"). Which commands run as which user on the hub host, and which need its root.
2. Re-pin, rollback and backup on the hub host, and who runs them after the deploy key is gone.
3. How the hub gets model access without a copy of jwoltje's `claude-session.env` or any factory credential.
If that needs the credentials row (T1 item 6), say so and name the interim. If it needs a new subscription
or a new provider account, flag it for Jason and don't pick one (mos-1008).
4. The books repo path. Which git.mosaicstack.dev remote, which credential the hub holds for it, and that it
can't read or push any other repository.
5. The admin path and the end state. The host ends with no key the factory uid can read. The design names the
admin path that remains (for example the hypervisor console, or a key kept off the factory uid) and who
holds it.
6. The acceptance script, as code Duckworth can read. It takes Gizmoduck's list of infrastructure gate items
as its input (mos-1011), not a fixed set of checks. Each item names a path, its closed form and its
accepted form, and the script reports which form holds for each, or fails. The list covers at least:
- federation R8's separate-host form: the hub's bus socket, data root and config dir don't exist on the
factory host, the factory host mounts nothing of the hub's, and no factory agent holds a login or key
for the hub host
- B7: from the factory uid, `sudo -n`, `doas -n`, `su`, polkit, and `ssh -o BatchMode=yes` to the hub
host all fail
- the hypervisor and backup paths ops-01 measured (mos-1011: one PVE node the factory key has root on,
PBS on that node with unencrypted backups, a flat /16 LAN), each closed or accepted by Jason
7. The identity provider and TLS front end (B8, mos-1008, mos-1011). Either the factory holds no credential
that can create users, change groups, issue tokens or impersonate in the issuer the hub trusts, or the hub
uses an issuer the factory can't administer. A brand, tenant or app inside the same Authentik does not
count as separate. The same holds for the TLS front end that serves the hub. The proof is a failed admin
call from every token the factory's root can read. No credential value is printed in that proof.
1. The install as input values to C2-13's scripts: user, home, business file, roles, the human id, the
token slot path. Which commands run as `mosaic` through `sudo -u mosaic -H`, and which as root through
ops-01.
2. Where Case 1 differs from Case 2: the factory roles in place of the library business, no library sync,
and any precheck item that doesn't apply, each with a reason.
3. Model access through the existing factory account (Q185). Placing the credential on .53 is a step under
the Q173 secrets rule, done in the window. It is never copied by the script, and its value is never
printed.
4. Re-pin, rollback and backup on .53, and who runs them.
5. The acceptance checks, as code Duckworth can read. Each check names a path and gives it one of three
marks: closed, accepted under Q182 or Q183, or "not named in Q182 or Q183". An accepted path doesn't
fail the row. An unnamed one is reported to Mos, who takes it to Jason (the R8 N1 rule). The functional
checks fail the row when they fail: the unit runs as `mosaic`, the Console answers on its port on the
internal network only, and one tester message gets a reply.
## Boundaries
- Provisioning, hypervisor access and any infra or k3s change are Jason's (Q173 exclusions). Mos takes the
hypervisor residual (factory root on the PVE nodes, Q172) to Jason. It doesn't block this row.
- Melanie's server (`[email protected]`) is reference only, read only, under Jason's authorization in
mos-0995. Nothing on it changes, and her Hermes bots keep running.
- No tunnel, Traefik or other exposure outside the internal network. That waits for Jason, after Q175 SSO.
- No credential value is printed, logged or committed. Keys made for the deploy are removed at the end, and
the acceptance script proves it.
- The admin-call proof in point 7 and any call to the issuer or the TLS front end run only in a window Mos
approves, never as a test or an anonymous probe (Q175).
- Tests in the repository use scratch roots and stand-ins. The live deploy and the acceptance run on the hub
host are a window with a backup first (where there is state) and a STATE report after.
- Infra and k3s changes, exposure outside the internal network, and spend stay Jason's (Q173 exclusions).
- No change to dragon-lin's `~/.mosaic-dev`, `~/.config/mosaic-dev`, its live units or `~/.mosaic`.
- No credential value is printed, logged or committed.
- Tests use scratch roots and stand-ins. Nothing in a test reaches .53.
- The run on .53 is a window with Jason's or Mos's go, a backup first where state exists, and a STATE report
after.
## Gate
Duckworth approves DESIGN.md, then the manifest. Any repository change is green on Sage's gate (cli suite and
every `scripts/test-*.sh`), with the design's mutants killed. The row is done when the acceptance script passes
on the hub host from the factory uid, Duckworth reads its output, and Mos confirms.
every `scripts/test-*.sh`), with the design's mutants killed. The row is done when the window's acceptance
run passes on .53, Duckworth reads its output, and Mos confirms.
@@ -0,0 +1,70 @@
# Row 193 build rows: 193a, 193b1, 193b2, 193c and 193d (#1691)
Sage, 2026-10-20, for Mos's go mos-1034 item 4. Row 193's design is Morgana's
`agents/morgana/work/queue-193/DESIGN.md` at e2d161e2e on `morgana/row193` (r5, sha256
dd7fcb419ed8510e9d8d74b569183386d0942af8b68eecc0bbb5e182e677b523), approved by Duckworth (duckworth-0018).
Section 8 lists the tests and mutants, and section 9 the split. The brief behind them is
`docs/plans/2026-10-20_human-users.md`. Vikunja P19 #414/#415, Gitea #1691.
Owner Morgana for every row, reviewer Duckworth, gate owner Sage. Each row chains on its parent's candidate
(Q177) and rebases onto the parent's landing. Seats push only their own `<seat>/*` branches (mos-0981).
Single test files run freely, and suites and mutant batches need Sage's slot.
Boundaries for every row: no change to `~/.mosaic-dev`, `~/.config/mosaic-dev`, the live units, slot files, any
live access.json or `~/.mosaic`. No request to the live Console, mosaic.woltje.com or auth.diversecanvas.com.
No user thread goes live before 193b1's message filter has landed (mos-1021 B2, FED R8).
The owner-visible records are fixed by Mos (mos-1023 item 2) and named in one user sentence, which FED R6
reuses word for word: "The business owner sees your thread's agent's decisions, inbox entries, claims, tasks
and project files, but not your messages."
## Brief: 193a, the access level check widened to viewer
Chained on FED R1. Started early on `morgana/row193a` (508ee4d61) against R1's interface as the FED brief
names it, with Sage's go (sage-1373). R1's level check accepts `viewer` and still refuses any other new
level. Tests in `packages/business/tests/` (section 8, Business), with the four mutants in Morgana's
BUILD.md. FED R4 chains on this row.
Duckworth's N1 on R1 (duckworth-0033, sage-1417): with viewer added, `covers` called with no level would let a
viewer grant cover a collaborator ask. This row makes `level` required in `covers`, with a test and a mutant that
makes it optional again. R3 and R4 call `covers` with a level.
Gate: Duckworth approves the manifest. The business suite and every `scripts/test-*.sh` are green on Sage's
gate, and the mutants are killed.
## Brief: 193b1, viewer verbs, board refusals and the message filter
Chained on FED R3, FED R4 and row 196's verifier. The bus gets the viewer verb list, per-grant level
checks, board reads refused to users, and the section 5.7 `trail` and `replies` filter with `#readable`
owner-only. No schema change. The governance pin test names each item in the user sentence above. Bus tests
and mutants 1, 3, 6, 10, 11 and 12 from section 8.
Gate: Duckworth approves the manifest. The bus suite and every `scripts/test-*.sh` are green on Sage's gate,
and the mutants are killed.
## Brief: 193b2, thread shares (schema 3j)
Chained on 193b1. `thread_shares`, the share verbs, the share rules of `#readable` and `#threadRead`, and the
section 5.5 share filters. Schema 3i to 3j. The table is append-only, and row 181's restore guard and
`COUNTED` include it. Bus tests for shares and the migration, and mutants 2, 4, 7, 8 and 9 from section 8.
Going live needs a `--migrate` window with Jason. Sage takes it to Mos when this row is near.
Gate: Duckworth approves the manifest. The bus suite, the cli suite and every `scripts/test-*.sh` are green on
Sage's gate, and the mutants are killed.
## Brief: 193c, mosaic access user verbs and --level viewer
Chained on FED R5. `mosaic access user add/remove` and `grant --level viewer`, only where R5 lacks them. They
refuse tester and agent callers, write a receipt line and call `access.reload` (section 8, CLI). If R5
already has all of them, this row records that and closes with no code.
Gate: Duckworth approves the manifest. The cli suite and every `scripts/test-*.sh` are green on Sage's gate,
and the mutants are killed.
## Brief: 193d, the Console's shared list, share controls and viewer read-only
Chained on FED R6, row 192 and 193b2 (for share controls). A viewer has no composer or share control, and
with each UI check removed the bus still refuses (section 8, Webui).
Gate: Duckworth approves the manifest. The webui suite and every `scripts/test-*.sh` are green on Sage's
gate, and the mutants are killed.
+23 -3
View File
@@ -33,7 +33,7 @@ Gaps found while working go to `docs/plans/DEFERRED.md`, not here.
<!-- mosaic-queue:begin -->
Generated from `docs/plans/queue.json` revision 2247 by `queue render`. Do not edit between the markers; change the queue with `scripts/mosaic queue`.
Generated from `docs/plans/queue.json` revision 2361 by `queue render`. Do not edit between the markers; change the queue with `scripts/mosaic queue`.
| # | Piece | Owner | Issues | State | After | Gate | Brief | Note |
|---|---|---|---|---|---|---|---|---|
@@ -225,9 +225,29 @@ Generated from `docs/plans/queue.json` revision 2247 by `queue render`. Do not e
| 186 | stack-release --plan: the backup line says what the run will do (P19 404, row 185 N1/N2) | darkwing; reviewers filbert | #1684 | done | — | Filbert approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the three brief mutants killed (sage) | `docs/plans/2026-10-20_plan-backup-line.md` § Brief: the activate plan's backup line says what the run will do | Landed 269134d8d (author Darkwing) on 3c5da3a91. Patch-id c97b9f6e25cd matches the r1 candidate (manifest 449e9121, Filbert approve 29419). Sage gate on b367f5730+patch 09:00:20Z-09:08:59Z: node 2681/0, scripts 422/0, queue 156, network removed; 3c5da3a91 adds only a SESSIONS line. Not live: the stack runs copy 490e2771d3d0. Follow-ups on P19 404: Filbert M6 gap, store-unclean reading. |
| 187 | stack-release activate carries the row 178 proof pair (mos-0928 Part B A0) | darkwing; reviewers filbert | #1685 | done | 186 done | Filbert approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the brief mutants killed (sage) | `docs/plans/2026-10-20_activate-proof-pair.md` § Brief: activate carries the row 178 proof pair | Landed 9a71e4e80 (author Darkwing) on d17328a55; patch-id 6b57c47bc661 matches candidate 52a5ecc1d143 and Filbert's r1 reading. Sage gate passed 10:21:55Z: cli 446/446 and every package and scripts/test-*.sh green. Filbert approve 29424, rev 2204; follow-ups N1-N4 filed on P19. |
| 188 | stack-release tests: a carried proof pair and a proof file inside the repository (P19 409, 410; gates Part B) | darkwing; reviewers filbert | #1686 | done | 187 settled | Filbert approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the brief mutants killed (sage) | `docs/plans/2026-10-20_proof-test-gaps.md` § Brief: tests for the two proof-file paths row 187 left open | Landed 8fc0ee216 (author Darkwing) on 23a609c5c; patch-id c9a56b167c07 matches candidate 86aa549bad75 and Filbert's r1 reading; stack-release.sh blob 2586a2a37cd2 unchanged. Sage gate passed 10:58Z: cli 453/453 and every package and scripts/test-*.sh green. Filbert approve 29433, rev 2212; N1 (REPO logical path) to P19 after Part B (mos-0959). |
| 189 | stack-release activate --no-send drops the Console's --send and keeps the pointer in step (Q175 interim, mos-0972, P19 413) | sage; reviewers dewey | #1687 | in-review, round 1, request posted | 163 done | Dewey approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the brief mutants killed (sage) | `docs/plans/2026-10-20_activate-no-send.md` § Brief: activate --no-send drops the Console's --send | — |
| 190 | several Mosaic Stack instances on one host: per-instance config, data root, units and ports (MVP T1.1, mos-0974, Vikunja #414) | darkwing; reviewers filbert | #1688 | in-progress | — | Filbert approves DESIGN.md then the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the design mutants killed (sage) | `docs/plans/2026-10-20_multi-instance.md` § Brief: several Mosaic Stack instances on one host | — |
| 189 | stack-release activate --no-send drops the Console's --send and keeps the pointer in step (Q175 interim, mos-0972, P19 413) | sage; reviewers dewey | #1687 | done | 163 done | Dewey approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the brief mutants killed (sage) | `docs/plans/2026-10-20_activate-no-send.md` § Brief: activate --no-send drops the Console's --send | Landed ace27dc5e on refactor (cherry-pick of candidate d6116771f onto a4a42dbac). Gate GATE_PASSED on 0c177e9ef (base 0217b71c8), 2026-10-20T19:09:37Z-19:20:36Z, 25 suites rc=0. patch-id ce89378718f6 for both 0c177e9ef and ace27dc5e; range-diff differs only in the message. Landed manifest hashes to 67af69bd (Dewey approve 29485). Base drift 0217b71c8..a4a42dbac: 7 doc/README files, none on manifest paths, no test reads them. Brief re-pinned to blob 29a9e6d2d. |
| 190 | several Mosaic Stack instances on one host: per-instance config, data root, units and ports (MVP T1.1, mos-0974, Vikunja #414) | darkwing; reviewers filbert | #1688 | in-review, round 1, request posted | — | Filbert approves DESIGN.md then the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the design mutants killed (sage) | `docs/plans/2026-10-20_multi-instance.md` § Brief: several Mosaic Stack instances on one host | — |
| 191 | one kill switch for the bus, every run, Discord and new launches, from the CLI and the Console (MVP T1.4, mos-0974, Vikunja #414) | rocko; reviewers dewey | #1689 | in-progress | — | Dewey approves DESIGN.md then the manifest; cli, bus, discord and webui suites and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_kill-switch.md` § Brief: one kill switch for the bus, every run, Discord and new launches | — |
| 192 | Q175 phase 1: the Console verifies SSO itself (--sso), OIDC RP with oauth4webapi 3.8.8, no live change (MVP T1.2, #412) | launchpad; reviewers webby | #1690 | in-review, round 2, request posted | — | Webby approves the manifest; webui and cli suites, scripts/test-config.sh and every scripts/test-*.sh green on Sage's gate; design section 9 mutants killed (sage) | `docs/plans/2026-10-20_sso-phase1.md` § Brief: Q175 phase 1, the Console verifies SSO itself (--sso) | — |
| 193 | many humans per business, human roles, private and shared threads, one access.json store (MVP T1.3, Vikunja #414/#415) | morgana; reviewers duckworth | #1691 | in-progress | — | Duckworth approves DESIGN.md then the manifest; bus, business, cli and webui suites and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_human-users.md` § Brief: many humans per business, human roles, private and shared threads | — |
| 194 | agent lifecycle from the Console and the CLI (create, launch, terminate, regenerate, delete) with a per-agent hire/fire veto (MVP T1.5, Vikunja #414/#415) | honker; reviewers webby | #1692 | in-progress | — | Webby approves DESIGN.md then the manifest; bus, cli, business and webui suites and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_agent-lifecycle.md` § Brief: agent lifecycle from the Console and the CLI, with a per-agent hire/fire veto | — |
| 195 | row 195: the Case 1 install on 10.1.1.53 as mosaic under /srv/mosaic, reusing C2-13's scripts by input under row 190's canon (Q181, Q182/Q183, Q185; ops-01 root steps under #423 step 1) | darkwing; reviewers duckworth | #1693 | briefed | 190 done | Duckworth approves DESIGN.md then the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_hub-own-uid.md` § Brief: the Case 1 install on 10.1.1.53 (row 195) | sage-1439: rewritten to Case 1 on .53 (f24f2cd40). Chains on C2-13's candidate too once C2-13 is queued. Darkwing starts DESIGN.md after row 190's mutants. |
| 196 | the human boundary across uids and humans: Console, tester listener, broker.sock and the path-bound human proof (P19 333, before federation R3, mos-0989) | gander; reviewers duckworth | #1694 | in-progress | — | Duckworth approves DESIGN.md then the manifest; bus, webui and cli suites and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_human-boundary.md` § Brief: the human boundary across uids and humans (P19 333) | — |
| 197 | a second gate lane: two full gates at once on dragon-lin, design first (mos-0977 item 6) | herb; reviewers webby | #1695 | in-progress | — | Webby approves DESIGN.md then the manifest; cli suite and every scripts/test-*.sh green on Sage's gate, singly and in the two-lane proof (sage) | `docs/plans/2026-10-20_second-gate-lane.md` § Brief: a second gate lane, two full gates at once on dragon-lin | Order with row 198 (Sage, mos-1040 N1): 197 lands first. After B1 the shared tests are packages/webui/tests/serve.test.mjs:100 and preview.test.mjs:208. Row 198 rebases onto 197's landing before its manifest review. |
| 198 | retire the control board writer on 7331 (unauthenticated loopback writer, Q175; mos-1008) | bushroot; reviewers webby | #1696 | in-progress | 192 done, 197 done | Webby approves INVENTORY.md then the manifest; control-board, webui and cli suites and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_retire-control-board.md` § Brief: retire the control board writer on 7331 | Order with row 197 (Sage, mos-1040 N1): 197 lands first. Row 198 rebases onto 197's landing before its manifest review; shared tests are packages/webui/tests/serve.test.mjs:100 and preview.test.mjs:208. |
| 199 | MVP T1 item 6: a credential store per agent, own git identity, pushes to a remote (G11, G13; mos-0974) | megavolt; reviewers duckworth | #1697 | in-progress | — | Duckworth approves DESIGN.md; Mos rules on the authority model; Duckworth approves the manifest; bus, business, cli and tasks suites and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_agent-credentials.md` § Brief: a credential store per agent, own git identity, pushes to a remote | — |
| 200 | stack-release.sh activate --sso for the Q175 phase 2 window (row 192 F3; chained on 192) | launchpad; reviewers webby | #1698 | in-review, round 2, request posted | — | Webby approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_stack-release-sso.md` § Brief: stack-release.sh activates a Console with --sso (Q175 phase 2 prerequisite) | — |
| 201 | FED R1: access.json loader and validator, and covers (#414) | gizmoduck; reviewers duckworth | #1699 | in-review, round 1, request posted | — | Duckworth approves the manifest; business suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_federation-r1-r2-r5.md` § Brief: FED R1, the access.json loader and validator | — |
| 202 | FED R5: mosaic access list, grant and revoke (#414) | quackerjack; reviewers duckworth | #1700 | in-review, round 1, request posted | — | Duckworth approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_federation-r1-r2-r5.md` § Brief: FED R5, mosaic access list, grant and revoke | Chained on row 201's candidate (Q177), not its done: claimable now, lands after 201 and rebases onto its landing (Sage) |
| 203 | FED R2: the federation key in the system config (#414) | quackerjack; reviewers webby | #1701 | briefed | — | Webby approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_federation-r1-r2-r5.md` § Brief: FED R2, the federation key in the system config | Chained on row 192's candidate (Q177): starts on 192's candidate, lands after 192 (Sage) |
| 204 | 193a: access level check widened to viewer (#1691) | morgana; reviewers duckworth | #1702 | in-review, round 1, request posted | — | Duckworth approves the manifest; business suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_human-users-build.md` § Brief: 193a, the access level check widened to viewer | Chained on row 201's candidate (Q177), not its done: claimable now, lands after 201 and rebases onto its landing (Sage) |
| 205 | 193b1: viewer verbs, board refusals and the message filter (#1691; also after FED R3 and R4 when filed) | morgana; reviewers duckworth | #1703 | queued | 196 done | Duckworth approves the manifest; bus suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_human-users-build.md` § Brief: 193b1, viewer verbs, board refusals and the message filter | — |
| 206 | 193b2: thread shares, schema 3j (#1691; live needs a --migrate window with Jason) | morgana; reviewers duckworth | #1704 | queued | 205 done | Duckworth approves the manifest; bus suite, cli suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_human-users-build.md` § Brief: 193b2, thread shares (schema 3j) | — |
| 207 | 193c: mosaic access user verbs and --level viewer (#1691) | morgana; reviewers duckworth | #1705 | queued | 202 done | Duckworth approves the manifest; cli suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_human-users-build.md` § Brief: 193c, mosaic access user verbs and --level viewer | — |
| 208 | 193d: Console shared list, share controls and viewer read-only (#1691; also after FED R6 when filed) | morgana; reviewers duckworth | #1706 | queued | 192 done, 206 done | Duckworth approves the manifest; webui suite and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_human-users-build.md` § Brief: 193d, the Console's shared list, share controls and viewer read-only | — |
| 209 | FED R8: the hub window on 10.1.1.54, a recorded acceptance (#414, r4.11 b93f098f0; live window run by Jason and Mos, after the Case 2 install #416, Bean and Chet live and both Hermes gateways stopped) | sage; reviewers duckworth | — | queued | — | Every entry check's output is in the R8 record with who ran it; Jason signs in to the hub and sends one message, Melanie revokes, his next request is refused; Duckworth checks the record (sage) | `docs/plans/2026-10-20_federation-r1-r2-r5.md` § Brief: FED R8, the hub window on 10.1.1.54 | Stays queued until its entry gates near: Case 2 install on .54 (#416, by 2026-10-27), Bean and Chet live, both Hermes gateways on .54 stopped, R7, #412 phase 2, #419 answer. Filed per mos-1073 after duckworth-0034 (Sage) |
| 210 | Agent host access with sudo through the credential store: grant per agent and host, the stack does ssh and sudo, the agent never sees the key (#1707, Vikunja #424, Q183/Q184, due 10-27; chained on row 199's candidate) | gizmoduck; reviewers duckworth | #1707 | in-progress | — | Duckworth approves DESIGN.md, then the manifest; bus, business, cli suites and every scripts/test-*.sh green on Sage's gate; the mutants killed (sage) | `docs/plans/2026-10-20_host-credentials.md` § Brief: agent host access with sudo through the credential store (#424) | Design runs before 199 lands (Q177 chain on 199's candidate). after cleared so Gizmoduck can claim; lands only after 199 and rebases onto it. |
| 211 | Force-push guard: scripts/git-hooks/pre-push refuses non-fast-forward and delete pushes from every worktree, and scripts/slot-busy.sh exits 1 on BUSY (#1709, Vikunja #426, mos-1078/1083) | rocko; reviewers dewey | #1709 | briefed | — | Dewey approves the manifest; the new or changed scripts/test-*.sh and every other scripts/test-*.sh green on Sage's gate; the mutants killed; hook installed in the canonical checkout with a scripted refused force-push (sage) | `docs/plans/2026-10-20_force-push-guard.md` § Brief: a pre-push hook that refuses force-pushes, and a busy check that fails on BUSY (#426) | — |
<!-- mosaic-queue:end -->
+4263 -14
View File
File diff suppressed because it is too large Load Diff
-7
View File
@@ -215,13 +215,6 @@ an operator inspect content and delivery evidence after an uncertain result.
| `trail` | `{subject}`: task ref, decision UUID or message UUID. Linked records in broker write order. |
| `decisions` | `{list}`: `roles` or `recent`; human and reader sessions only. Open decisions routed to a role, or the last 50 closed. |
| `credentials` | No args; human and reader sessions only. `[{service,account,role,dateKind,date,state}]` for the capability's business, never a token, file path or variable name. |
| `credentials.issue` | `{agent,service,token,date}`; human only (row 199, #1697). Writes the token to `<configDir>/secrets/agents/<business>/<agent>/<service>` (0600, dirs 0700: tmp, fsync, rename), rewrites the agent's reference in the business file to `{"store":true,<rotateBy or expires>:date}`, then holds the token in memory, with no restart. Over an issued entry it is a rotate: the old value is gone from disk and memory when it returns. `service` is `gitea` or `vikunja` and the business file must already name it for the agent (`credential-unknown`). A token the bus already holds is `credential-leak`; a bad token `credential-format`; a bad date `credential-date`. Returns `{agent,verb,service,date,state}`, `verb` `issue` or `rotate`. The tester gets `tester-refused`. |
| `credentials.revoke` | `{agent,service?,reason}`; human, or the tester (Q173). Drops the entry from memory, then unlinks a store file; a `file` or `env` reference keeps its file and becomes `store`, so a restart loads nothing. Without `service`, every kind the agent has. Returns `{agent,verb,services:[{service,state}]}`. |
| `credentials.delete` | `{agent,reason}`; human only. `credentials.revoke` for every kind, then removes the agent's store dir; a file there other than a token is `credential-file`. While the agent or one of its threads holds a claim, `agent-running`. |
Each of the three leaves one receipt in the organization's `changes.jsonl`: the actor and a
`detail` of `{verb,agent,kind,date or reason}`, never a value. A change to the store alone writes
a receipt with empty `changes`. A refusal is `action.refused` with its code alone.
| `requirements` | No args; role sessions only, as `task.create` is (row 90, #1584). The approved PRD's requirement ids, `{prd,version,requirements:[{id,title}]}`, from `packages/tasks/requirements.json`, loaded when the broker starts; a bad file refuses the start. A human session refuses `agent-required`, a reader `read-only`. |
| `state` | No args; role, human and reader sessions (row 109, #1603). `{launch:{state,at,event},tasks:{open:[{task_ref,title}],total,cap}}` for the capability's business. `launch` is the latest `launch.revoked` or `launch.restored` event: `allowed` with `at` and `event` null when none exists. `open` is the tracker snapshot's open tasks in `task_ref` order, at most `cap` (the `tasks` page size, 50); `total` counts every open task. The harness runner reads it before each turn. Writes nothing. |
| `replies` | `{id}`: a message UUID; human and reader sessions only. The messages whose `in_reply_to` is that id, oldest first, at most 50. Writes nothing. |
+6 -153
View File
@@ -3,7 +3,7 @@ import { isAbsolute } from 'node:path';
import { NAME as HARNESS_NAME } from '../../plugins/src/kinds.mjs';
import { loadRequirements } from '../../tasks/src/requirements.mjs';
import { MERGE, LandingRefusal, readLanding } from './landing.mjs';
import { ID_PATTERN, THREAD_DEFINITION, END_PROTECTED, forkProblem, forkUrl } from '../../business/src/vocabulary.mjs';
import { ID_PATTERN, THREAD_DEFINITION, END_PROTECTED } from '../../business/src/vocabulary.mjs';
// Kept closed while S1 lands; its vocabulary is the integration authority.
export const ACTIONS = Object.freeze([
@@ -18,8 +18,6 @@ export const ACTIONS = Object.freeze([
'git.push.working',
'git.push.protected',
'git.merge.protected',
// Row 199 (#1697, mos-1027 Q-M2): publish to the agent's fork. Not in ROUTED, so a hard wait.
'git.push.remote',
'review.request',
'review.verdict',
'record.write',
@@ -103,11 +101,6 @@ export const TESTER_VERBS = Object.freeze(['message.send', 'message.receive', 'm
// Row 175: the organization verbs a tester may call. Its writes run every check the human's do
// and log actor `tester`; row 176 adds routes reads; limits.set, organization.set and organization.revision refuse it.
export const TESTER_ORGANIZATION_VERBS = Object.freeze(['limits', 'plugins', 'instances', 'plugins.set', 'instances.set', 'routes.set', 'routes']);
// Row 199 (#1697, design section 6): the credential lifecycle, human-only. Under Q173 the tester may
// revoke; issue (rotate is an issue over an issued entry) and delete refuse it, because a token it
// issued would pass through its session (duckworth N1). The status read stays the `credentials` verb.
export const CREDENTIAL_VERBS = Object.freeze(['credentials.issue', 'credentials.revoke', 'credentials.delete']);
const TESTER_CREDENTIAL_VERBS = Object.freeze(['credentials.revoke']);
const SELF_KINDS = new Set(['task.created', 'task.assigned', 'task.state', 'task.closed', 'task.conflict']);
const POLL_KINDS = new Set([
'task.changed.external',
@@ -198,13 +191,10 @@ export class Broker {
#received = new Set();
// Row 137 (#1631): landings checkLand let start, by ticket, until recordLand or landRefused ends them.
#landings = new Map();
// Row 199: a publish between checkPublish and recordPublish or publishRefused.
#publishes = new Map();
#secretCheck;
#credentialStatus;
#requirements;
#organization;
#credentialVerbs;
#project;
#usage;
#limits;
@@ -213,15 +203,13 @@ export class Broker {
#stops = new Map();
#clock = 0;
// credentialStatus is the runtime's metadata-only Credentials.status(); the broker never gets use().
constructor({ store, businesses, secretCheck = () => {}, credentialStatus = () => [], organization = null, project = null, usage = null, limits = null, harnessOn = null, credentialVerbs = null }) {
constructor({ store, businesses, secretCheck = () => {}, credentialStatus = () => [], organization = null, project = null, usage = null, limits = null, harnessOn = null }) {
if (!object(businesses) || !Object.keys(businesses).length) fail('invalid-business');
this.#businesses = structuredClone(businesses);
this.#store = store;
this.#secretCheck = secretCheck;
this.#credentialStatus = credentialStatus;
this.#organization = organization;
// Row 199: the runtime's credential lifecycle (credential-verbs.mjs); it holds the store, not the broker.
this.#credentialVerbs = credentialVerbs;
this.#project = project;
this.#usage = usage;
this.#limits = limits;
@@ -807,12 +795,7 @@ export class Broker {
// records `?? null` and the current holder reads undefined.
if (action === 'role.revoke' && (context.holderRun ?? null) !== (this.#holder(s.business, target)?.holder_run ?? null))
fail('decision-mismatch');
// Row 199 (r4 B4): approve at A, push B, publish refuses.
if (action === 'git.push.remote' && (context.pushCommit ?? null) !== (this.#latestPush(s.business, target)?.commit ?? null))
fail('decision-mismatch');
return action === 'git.push.remote'
? { class: cls, decision: d.id, pushCommit: context.pushCommit ?? null }
: { class: cls, decision: d.id };
return { class: cls, decision: d.id };
}
// The raise's context: its target and approval option, from the decision.raise action.allowed.
#raised(business, decision) {
@@ -875,14 +858,13 @@ export class Broker {
if (body.commit !== push.commit) fail('verdict-stale');
}
// A push is the action.allowed recordPush wrote, the only one for git.push.working with a commit.
// role is the role that pushed (row 199, work-not-yours).
#latestPush(business, task) {
const r = this.#store.get(
"SELECT body, actor_role FROM events WHERE business=? AND kind='action.allowed' AND json_extract(body,'$.action')='git.push.working' AND json_type(body,'$.commit')='text' AND subject=? ORDER BY seq DESC LIMIT 1",
"SELECT body FROM events WHERE business=? AND kind='action.allowed' AND json_extract(body,'$.action')='git.push.working' AND json_type(body,'$.commit')='text' AND subject=? ORDER BY seq DESC LIMIT 1",
business,
task,
);
return r ? { ...JSON.parse(r.body), role: r.actor_role } : null;
return r ? JSON.parse(r.body) : null;
}
// Row 93: the bus host's git socket asks before it takes a push's bytes. Nothing is recorded
// unless it refuses: the push is recorded, and its authority consumed, once the ref moved.
@@ -1016,98 +998,6 @@ export class Broker {
this.#landings.delete(ticket);
return code;
}
// Row 199 (#1697, design D9 steps 1 to 3, D9b): the bus host's git socket publishes a task's
// recorded work commit to the agent's own fork. checkPublish runs before anything touches the
// network. The run holds its claim; git.push.remote is a hard wait, so it cites a decision the
// human resolved for this task, bound to the task's latest push (#authorityFor). The project must
// be in the agent's publish grant with an upstream; the fork is built from that upstream and the
// agent's own Gitea user, never from the request. The ticket carries the session and decision to
// recordPublish or publishRefused. No token is in the result.
checkPublish(cap, args) {
const s = this.#session(cap);
return this.#store.transaction(() => {
keys(args, ['task_ref', 'project', 'branch', 'decision'], ['task_ref', 'project', 'branch']);
if (!taskRef(args.task_ref) || (args.decision !== undefined && !id(args.decision))) fail('invalid-request');
identifier(args.project);
string(args.branch, 200);
this.#agent(s);
const { task_ref, project, branch } = args;
const decision = args.decision ?? null;
const result = this.#checkAuthority(s, 'git.push.remote', { decision, target: task_ref });
const facts = this.#publishFacts(s, task_ref, project, result);
const ticket = randomUUID();
const dst = `refs/heads/${facts.user}/${branch}`;
this.#publishes.set(ticket, { s, task_ref, project, decision, dst, ...facts });
return { ticket, task_ref, project, fork: facts.fork, user: facts.user, email: facts.email, source: facts.source, dst };
});
}
// Steps 2 and 3, checked again at recordPublish. After routing a within-role publish has no
// decision, so the commit is the latest push's (N9); the git socket still checks the work ref tip.
#publishFacts(s, task_ref, project, result) {
if (result.decision && this.#spent(s.business, result.decision)) fail('decision-consumed');
const b = this.#business(s.business);
const role = b.roles?.[s.role];
const upstream = b.projects?.[project]?.remote;
if (!role?.identity || !(role.publish ?? []).includes(project) || !upstream) fail('remote-not-granted');
const problem = forkProblem(upstream, role.identity);
if (problem) fail(problem);
const push = this.#latestPush(s.business, task_ref);
if (!push) fail('no-push');
if (push.role !== s.role) fail('work-not-yours');
if (push.project !== project) fail('work-ref');
const source = result.pushCommit ?? push.commit;
if (source !== push.commit) fail('decision-mismatch');
const user = role.identity.gitea.user;
return { fork: forkUrl(upstream, user), user, email: role.identity.email, source };
}
#publish(ticket) {
return (typeof ticket === 'string' && this.#publishes.get(ticket)) || fail('publish-unknown');
}
// Design D5: the only key the git socket's gitea token use gets is the agent checkPublish checked
// (#agent: a run holding its claim, never a tester, reader or human). Trusted host only.
publishSession(ticket) {
const { s } = this.#publish(ticket);
return Object.freeze({ business: s.business, role: s.role });
}
// Where the bus process's git sends the push (D9 step 5): all from the ticket, none from the
// host's message. Trusted host only.
publishTarget(ticket) {
const { s, project, fork, user, source, dst } = this.#publish(ticket);
return Object.freeze({ business: s.business, project, fork, user, source, dst });
}
// Step 6: records the push the git socket made as the action.allowed for git.push.remote, which
// spends the decision. The checks run again; the run may have ended while git ran, so the holder
// check is not repeated (as recordLand). Trusted host only.
recordPublish(ticket, pushed) {
const t = this.#publish(ticket);
const done = this.#store.transaction(() => {
keys(pushed, ['ref', 'commit', 'previous'], ['ref', 'commit', 'previous']);
if (!oid(pushed.commit) || (pushed.previous !== null && !oid(pushed.previous))) fail('invalid-request');
string(pushed.ref, 260);
const result = this.#authorityFor(t.s, 'git.push.remote', { decision: t.decision, target: t.task_ref });
const facts = this.#publishFacts(t.s, t.task_ref, t.project, result);
if (facts.source !== t.source || pushed.commit !== t.source || facts.fork !== t.fork || pushed.ref !== t.dst) fail('ref-moved');
const { ref, commit, previous } = pushed;
this.#event(
t.s,
'action.allowed',
{ project: t.project, fork: t.fork, ref, commit, previous, action: 'git.push.remote', ...result, target: t.task_ref },
t.task_ref,
);
return { task_ref: t.task_ref, fork: t.fork, ref, commit, previous };
});
this.#publishes.delete(ticket);
return done;
}
// Ends a publish that didn't push: action.refused with the code on the task's trail. The git
// socket answers the run with the same code. Trusted host only.
publishRefused(ticket, { code } = {}) {
const t = this.#publish(ticket);
if (!/^[a-z-]{1,64}$/.test(code ?? '')) fail('invalid-request');
this.#store.transaction(() => this.#event(t.s, 'action.refused', { code }, t.task_ref));
this.#publishes.delete(ticket);
return code;
}
// The outcome as a RESULT from the bus, as #notice sends: the runner takes a turn on it and sends
// no automatic reply. It goes to the role in the thread the decision was raised in.
#landMessage(s, decision, body) {
@@ -1355,7 +1245,6 @@ export class Broker {
return this.#requestProject(s, { verb: request.verb, args: structuredClone(args) });
if (PROJECT_FILE_VERBS.includes(request.verb))
return this.#requestProjectFile(s, { verb: request.verb, args: structuredClone(args) });
if (CREDENTIAL_VERBS.includes(request.verb)) return this.#requestCredential(s, request.verb, structuredClone(args));
if (["limits", "limits.set", "organization.set", "organization.revision", "plugins", "plugins.set", "instances", "instances.set", "routes", "routes.set"].includes(request.verb)) {
// Row 123 (#1617, REQ-COS-1, REQ-LIMIT-1): organization.set is limits.set for a role whose
// file lists organization.change. It takes the same keys, runs the same write path and
@@ -1427,35 +1316,6 @@ export class Broker {
throw this.#refused(s, e, request);
}
}
// Row 199 (#1697, design section 6). The token of an issue comes in args from the human's stdin; the
// secret check above has refused one the bus already holds, a refusal records its code alone, and
// nothing returned carries it. Outside the store transaction, as the organization branch.
#requestCredential(s, verb, a) {
if (s.tester ? !TESTER_CREDENTIAL_VERBS.includes(verb) : !s.human) fail(s.tester ? 'tester-refused' : 'human-required');
if (!this.#credentialVerbs) fail('not-configured');
if (verb === 'credentials.issue') keys(a, ['agent', 'service', 'token', 'date'], ['agent', 'service', 'token', 'date']);
else keys(a, verb === 'credentials.revoke' ? ['agent', 'service', 'reason'] : ['agent', 'reason'], ['agent', 'reason']);
string(a.agent, 64);
if (a.service !== undefined && !['gitea', 'vikunja'].includes(a.service)) fail('invalid-request');
if (a.token !== undefined) string(a.token, 16384);
if (a.date !== undefined) string(a.date, 10);
if (a.reason !== undefined) string(a.reason, 500);
if (!Object.hasOwn(this.#business(s.business).roles, a.agent)) fail('credential-unknown');
// A running agent keeps its tokens until its runs end: delete waits for every claim to end.
if (verb === 'credentials.delete' && (this.#holder(s.business, a.agent)?.op === 'claim' || this.#threadHolders(s.business, a.agent).length))
fail('agent-running');
let result;
try {
result = this.#credentialVerbs(s.business, verb, a, s.tester ? 'tester' : s.human);
} catch (e) {
if (e instanceof BusError && e.code.startsWith('credential-')) throw e;
if (e.code === 'revision-conflict' || e.code?.startsWith('organization-')) fail(e.code);
if (e.name === 'BusinessError') fail('invalid-request');
throw new BusError('outcome-unknown');
}
this.#secretCheck(result);
return result;
}
// Only the runner's bound session and delivered message identify a turn; no host path crosses the wire.
// Row 156 (#1651, sage-0808): a role run's landing turn is a merge decision id instead. Its begin
// needs the decision held for the run's role; its end needs that begin, since the landing may
@@ -1838,12 +1698,6 @@ export class Broker {
by,
);
}
// Row 199 (#1697, design D9 step 1, r4 B4): a publish decision binds one commit, the task's
// latest recorded push now, read here and never taken from the agent. The human sees it.
const pushCommit =
a.action === 'git.push.remote' ? (this.#latestPush(s.business, a.target)?.commit ?? fail('no-push')) : null;
const question = pushCommit ? `${a.question} (commit ${pushCommit})` : a.question;
string(question);
const did = randomUUID();
this.#store.run(
'INSERT INTO decisions(id,at,business,project,raised_by_role,raised_by_run,class,action,route_to,question,options,recommendation,task_ref,requirement_ref,blocking,supersedes,spec_path,spec_commit) VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',
@@ -1856,7 +1710,7 @@ export class Broker {
cls,
a.action,
route,
question,
a.question,
JSON.stringify(a.options),
a.recommendation,
a.task_ref ?? null,
@@ -1878,7 +1732,6 @@ export class Broker {
approvalChoice,
holderRun:
a.action === 'role.revoke' ? (this.#holder(s.business, a.target)?.holder_run ?? null) : null,
...(pushCommit ? { pushCommit } : {}),
},
a.task_ref ?? (taskRef(a.target) ? a.target : did),
);
-53
View File
@@ -1,53 +0,0 @@
import { BusError } from './broker.mjs';
// Row 199 (#1697, design section 6): issue, rotate, revoke and delete of one agent's credentials, for
// the broker's human-only verbs. The broker checks the session, the keys and a running claim first.
// Each verb leaves one receipt in the organization's changes.jsonl: the actor and a detail naming the
// verb, agent, kinds, date or reason, never a value. Nothing here logs, and a result carries no token.
const DATE_KEY = Object.freeze({ gitea: 'rotateBy', vikunja: 'expires' });
const deny = (code) => {
throw new BusError(code);
};
export function credentialVerbs({ credentials, writer }) {
const state = (instance, service) =>
credentials.status().find((e) => e.instance === instance && e.service === service)?.state;
return (id, verb, args, actor) => {
const { agent } = args;
const instance = id + '/' + agent;
const { document, revision } = writer.read(id);
const current = document.roles?.[agent]?.credentials;
if (!current) deny('credential-unknown');
if (verb === 'credentials.issue') {
const { service, token, date } = args;
if (!current[service]) deny('credential-unknown');
// rotate is an issue over an entry the bus holds now; the old value is gone when this returns.
const kind = ['not-issued', 'revoked', undefined].includes(state(instance, service)) ? 'issue' : 'rotate';
const next = structuredClone(document);
next.roles[agent].credentials[service] = { store: true, [DATE_KEY[service]]: date };
const detail = { verb: kind, agent, kind: [service], date };
const status = credentials.issue(instance, service, token, date, () =>
writer.write(id, next, { revision, actor, detail }),
);
return { agent, verb: kind, service, date: status.date, state: status.state };
}
const services = verb === 'credentials.revoke' && args.service ? [args.service] : Object.keys(current);
if (services.some((service) => !current[service])) deny('credential-unknown');
// Memory first, then the store file (Credentials.revoke), then the business file. A reference
// outside the store becomes `store` with its date, so a restart loads nothing for it. When that
// write fails, the entry stays revoked in memory and the refusal says to run the verb again.
if (verb === 'credentials.delete') credentials.remove(instance);
else for (const service of services) credentials.revoke(instance, service);
const detail = { verb: verb.slice('credentials.'.length), agent, kind: services, reason: args.reason };
const moved = services.filter((service) => current[service].store !== true);
if (moved.length) {
const next = structuredClone(document);
for (const service of moved)
next.roles[agent].credentials[service] = { store: true, [DATE_KEY[service]]: current[service][DATE_KEY[service]] };
writer.write(id, next, { revision, actor, detail });
} else writer.record(id, { actor, detail });
return {
agent,
verb: detail.verb,
services: services.map((service) => ({ service, state: state(instance, service) ?? 'not-issued' })),
};
};
}
+18 -162
View File
@@ -1,8 +1,5 @@
import {
openSync, closeSync, fstatSync, readFileSync, lstatSync, realpathSync, constants, mkdirSync, writeSync, fsyncSync,
renameSync, unlinkSync, rmdirSync, readdirSync,
} from 'node:fs';
import { dirname, isAbsolute, join, relative, resolve } from 'node:path';
import { openSync, closeSync, fstatSync, readFileSync, lstatSync, realpathSync, constants } from 'node:fs';
import { isAbsolute, relative, resolve } from 'node:path';
import { BusError } from './broker.mjs';
const deny = (code) => {
throw new BusError(code);
@@ -16,33 +13,9 @@ const day = (value) =>
/^\d{4}-\d{2}-\d{2}$/.test(value) &&
!Number.isNaN(Date.parse(value)) &&
new Date(value).toISOString().slice(0, 10) === value;
const TOKEN = /^[-A-Za-z0-9._~]+\n?$/;
// Row 199 (#1697, design D1, D4): a store dir is the stack instance's user's own and nobody else's.
const privateDir = (path, { missing = false } = {}) => {
let st;
try {
st = lstatSync(path);
} catch (e) {
if (missing && e.code === 'ENOENT') return false;
deny('credential-file');
}
if (!st.isDirectory() || st.uid !== process.getuid() || (st.mode & 0o077) !== 0) deny('credential-file');
return true;
};
export class Credentials {
#tokens = new Map();
// Row 199: every reference the bus loaded, so issue, rotate, revoke and status work per entry.
#refs = new Map();
#revoked = new Set();
#storeRoot;
#repoRoots;
#dataRoot;
// storeRoot is <configDir>/secrets/agents (design D1); a reference with store: true must name the
// path under it for its own business, agent and service.
constructor({ references = {}, repoRoots = [], dataRoot, env = process.env, storeRoot = null }) {
this.#storeRoot = storeRoot;
this.#repoRoots = repoRoots;
this.#dataRoot = dataRoot;
constructor({ references = {}, repoRoots = [], dataRoot, env = process.env }) {
try {
for (const [instance, services] of Object.entries(references))
for (const [service, ref] of Object.entries(services)) {
@@ -53,41 +26,12 @@ export class Credentials {
if (
(ref.service !== undefined && ref.service !== service) ||
Object.keys(ref).some(
(k) => !['service', 'file', 'env', 'store', service === 'vikunja' ? 'expires' : 'rotateBy'].includes(k),
(k) => !['service', 'file', 'env', service === 'vikunja' ? 'expires' : 'rotateBy'].includes(k),
) ||
Boolean(ref.file) === Boolean(ref.env) ||
(ref.store !== undefined && (ref.store !== true || ref.file !== this.#storePath(instance, service)))
Boolean(ref.file) === Boolean(ref.env)
)
deny('credential-reference');
this.#refs.set(instance + ':' + service, { ref, date });
const token = this.#read(ref, env);
if (token !== null) this.#tokens.set(instance + ':' + service, { token, date });
}
} catch (e) {
this.close();
throw e;
}
}
#storePath(instance, service) {
if (!this.#storeRoot) deny('credential-reference');
const [business, agent, ...rest] = instance.split('/');
if (!business || !agent || rest.length || agent.startsWith('@')) deny('credential-reference');
return join(this.#storeRoot, business, agent, service);
}
// One token, read and checked, or null for a store reference with no file yet (design D3a).
#read(ref, env) {
let token;
if (ref.store) {
const agentDir = dirname(ref.file);
privateDir(this.#storeRoot);
if (!privateDir(dirname(agentDir), { missing: true }) || !privateDir(agentDir, { missing: true })) return null;
try {
lstatSync(ref.file);
} catch (e) {
if (e.code === 'ENOENT') return null;
deny('credential-file');
}
}
if (ref.file) {
if (!isAbsolute(ref.file)) deny('credential-location');
let resolved, original;
@@ -99,7 +43,7 @@ export class Credentials {
}
if (original.isSymbolicLink() || !original.isFile()) deny('credential-file');
if (
[...this.#repoRoots, this.#dataRoot].filter(Boolean).some((root) =>
[...repoRoots, dataRoot].filter(Boolean).some((root) =>
inside(
(() => {
try {
@@ -137,98 +81,16 @@ export class Credentials {
if (!/^[A-Z_][A-Z0-9_]*$/.test(ref.env)) deny('credential-reference');
token = env[ref.env];
}
return Credentials.#format(token);
}
static #format(token) {
if (typeof token !== 'string' || token.length > 16384 || !TOKEN.test(token)) deny('credential-format');
token = token.replace(/\n$/, '');
if (token.length < 16) deny('credential-format');
return token;
}
#known(instance, service) {
const known = this.#refs.get(instance + ':' + service);
if (!known) deny('credential-unknown');
return known;
}
// Row 199 (design section 6): write the token to the agent's store path (tmp, fsync, rename), then
// `commit` (the business file's reference becomes store), then memory. The token comes from the
// caller's stdin and is never logged. Returns the entry's new status.
issue(instance, service, token, date, commit = () => {}) {
this.#known(instance, service);
if (!day(date)) deny('credential-date');
token = Credentials.#format(token);
const file = this.#storePath(instance, service);
const agentDir = dirname(file);
mkdirSync(this.#storeRoot, { recursive: true, mode: 0o700 });
privateDir(this.#storeRoot);
for (const dir of [dirname(agentDir), agentDir]) {
if (!privateDir(dir, { missing: true })) mkdirSync(dir, { mode: 0o700 });
privateDir(dir);
}
const tmp = file + '.tmp';
try {
unlinkSync(tmp);
if (typeof token !== 'string' || token.length > 16384 || !/^[-A-Za-z0-9._~]+\n?$/.test(token))
deny('credential-format');
token = token.replace(/\n$/, '');
if (token.length < 16) deny('credential-format');
this.#tokens.set(instance + ':' + service, { token, date });
}
} catch (e) {
if (e.code !== 'ENOENT') deny('credential-file');
this.close();
throw e;
}
let fd;
try {
fd = openSync(tmp, constants.O_WRONLY | constants.O_CREAT | constants.O_EXCL | constants.O_NOFOLLOW, 0o600);
writeSync(fd, token + '\n');
fsyncSync(fd);
} finally {
if (fd !== undefined) closeSync(fd);
}
renameSync(tmp, file);
const dirFd = openSync(agentDir, constants.O_RDONLY);
try {
fsyncSync(dirFd);
} finally {
closeSync(dirFd);
}
commit();
this.#refs.set(instance + ':' + service, { ref: { service, store: true, file, [service === 'vikunja' ? 'expires' : 'rotateBy']: date }, date });
this.#tokens.set(instance + ':' + service, { token, date });
this.#revoked.delete(instance + ':' + service);
return this.status().find((e) => e.instance === instance && e.service === service);
}
// Memory first, then the store file (design section 6, D12). A file or env reference outside the
// store is never unlinked; the caller turns its reference into store, so a restart finds nothing.
revoke(instance, service) {
const { ref } = this.#known(instance, service);
this.#tokens.delete(instance + ':' + service);
this.#revoked.add(instance + ':' + service);
if (ref.store) {
try {
unlinkSync(ref.file);
} catch (e) {
if (e.code !== 'ENOENT') deny('credential-file');
}
}
return { file: Boolean(ref.store) };
}
// revoke for every service, then the agent's dir. A file other than a token or its tmp refuses.
remove(instance) {
const services = [...this.#refs.keys()].filter((k) => k.startsWith(instance + ':')).map((k) => k.slice(instance.length + 1));
for (const service of services) this.revoke(instance, service);
const agentDir = dirname(this.#storePath(instance, 'gitea'));
let names;
try {
names = readdirSync(agentDir);
} catch (e) {
if (e.code === 'ENOENT') return { services };
deny('credential-file');
}
if (names.some((n) => !/^(gitea|vikunja)(\.tmp)?$/.test(n))) deny('credential-file');
for (const n of names) unlinkSync(join(agentDir, n));
rmdirSync(agentDir);
return { services };
}
// The agent-facing use (design D5): the key comes only from the broker's verified session.
async useFor(session, service, operation) {
if (!session || typeof session.business !== 'string' || typeof session.role !== 'string' || session.role.startsWith('@'))
deny('credential-not-yours');
return this.use(session.business + '/' + session.role, service, operation);
}
assertClean(value) {
let encoded;
@@ -247,16 +109,14 @@ export class Credentials {
let result;
try {
result = await operation(entry.token);
} catch (e) {
// Row 199: a host step's own refusal (publish) keeps its code; anything else is the service's.
if (e instanceof BusError) throw e;
} catch {
deny('service-failed');
}
this.assertClean(result);
return result;
}
status() {
return [...this.#refs.entries()].map(([key, { date }]) => {
return [...this.#tokens.entries()].map(([key, { date }]) => {
const i = key.lastIndexOf(':'),
service = key.slice(i + 1),
left = Date.parse(date + 'T00:00:00Z') - Date.now();
@@ -264,11 +124,8 @@ export class Credentials {
instance: key.slice(0, i),
service,
date,
state: !this.#tokens.has(key)
? this.#revoked.has(key)
? 'revoked'
: 'not-issued'
: service === 'gitea'
state:
service === 'gitea'
? left <= 0
? 'rotation-due'
: 'valid'
@@ -282,6 +139,5 @@ export class Credentials {
}
close() {
this.#tokens.clear();
this.#refs.clear();
}
}
+1 -13
View File
@@ -14,11 +14,6 @@ const LAUNCH_OPS = new Set([
'checkLand',
'recordLand',
'landRefused',
'checkPublish',
'publishPush',
'publishResult',
'recordPublish',
'publishRefused',
'landHeld',
'threadsState',
'threadIdle',
@@ -39,12 +34,6 @@ function launchOp(m) {
if (m.op === 'checkLand') return runtime.checkLand(m.cap, m.args);
if (m.op === 'recordLand') return runtime.recordLand(m.ticket, m.land);
if (m.op === 'landRefused') return runtime.landRefused(m.ticket, m.args);
// Row 199 (#1697, design D9b): the git socket's publish op. No entry returns a token.
if (m.op === 'checkPublish') return runtime.checkPublish(m.cap, m.args);
if (m.op === 'publishPush') return runtime.publishPush(m.ticket);
if (m.op === 'publishResult') return runtime.publishResult(m.ticket);
if (m.op === 'recordPublish') return runtime.recordPublish(m.ticket, m.pushed);
if (m.op === 'publishRefused') return runtime.publishRefused(m.ticket, m.args);
if (m.op === 'landHeld') return runtime.landHeld(m.record);
// Row 121 (#1615): the launcher's view of the business's threads, and its idle release.
if (m.op === 'threadsState') return runtime.threadsState(m.business);
@@ -110,8 +99,7 @@ if (!process.send) {
booted = true;
clearTimeout(timer);
// `trackers` is S3's plain-data boot config; the adapter is loaded only when a business has one.
// Row 199: publishTls is the tests' scratch CA; no boot config sets it.
const { trackers, publishTls: _tls, ...config } = message.config ?? {};
const { trackers, ...config } = message.config ?? {};
if (trackers) {
const { tasksAdapter } = await import('../../tasks/src/adapter.mjs');
config.tasks = tasksAdapter({ trackers });
+1 -56
View File
@@ -8,12 +8,9 @@ import { Credentials } from './credentials.mjs';
import { serve } from './server.mjs';
import { launchSettings } from '../../business/src/launch-settings.mjs';
import { readSessions } from '../../seat/src/session.mjs';
import { publishRemote, workingRepo, WorkingError } from '../../seat/src/working.mjs';
import { createLoader, hasModule } from '../../plugins/src/loader.mjs';
import { pluginSettings } from '../../business/src/plugin-settings.mjs';
import { instanceSettings } from '../../business/src/instance-settings.mjs';
import { organizationWriter } from '../../business/src/organization.mjs';
import { credentialVerbs } from './credential-verbs.mjs';
import { webService } from './web.mjs';
import { verifyHuman } from './human.mjs';
import { createUsage } from './usage.mjs';
@@ -75,9 +72,6 @@ export async function startBroker({
repoRoots = [],
tasks = null,
organization = null,
// Row 199: the tests' scratch CA for publish's git (`-c http.sslCAInfo=...`). process.mjs drops
// it from the boot config, so only an in-process caller sets it.
publishTls = [],
}) {
let store, credentials, server, adapter;
try {
@@ -92,9 +86,7 @@ export async function startBroker({
...repoRoots,
...Object.values(businesses).flatMap((b) => Object.values(b.projects ?? {}).map((p) => p.root)),
];
// Row 199 (#1697, design D1): the per-agent store is <configDir>/secrets/agents.
const storeRoot = organization ? join(organization.dir, 'secrets', 'agents') : null;
credentials = new Credentials({ references, repoRoots: protectedRoots, dataRoot, storeRoot });
credentials = new Credentials({ references, repoRoots: protectedRoots, dataRoot });
store = new Store(dataRoot);
// vikunja:19/305: the plug-in switch through the same human-only path as the limits: the verbs
// plugins and plugins.set, logged in changes.jsonl with the actor, a missing module refused.
@@ -141,11 +133,6 @@ export async function startBroker({
}:null,
secretCheck: (value) => credentials.assertClean(value),
credentialStatus: () => credentials.status(),
// Row 199 (#1697, design section 6): issue, rotate, revoke and delete, live with no restart; each
// rewrites or records through the organization writer, so changes.jsonl holds the receipt.
credentialVerbs: organization
? credentialVerbs({ credentials, writer: organizationWriter({ dir: organization.dir, rolesDir: organization.rolesDir }) })
: null,
// Row 65: a new launch binds only on a harness its business switches on, read from the file
// as it stands now; an unreadable file switches none on.
harnessOn: (id, name) => {
@@ -225,42 +212,6 @@ export async function startBroker({
// Row 137 (#1631): the git socket's land op. checkLand's refusal is the run's, like checkPush's.
// A landing then ends in recordLand or landRefused under its ticket, which outlive the run's cap.
const checkLand = (cap, args) => refusing(cap, () => broker.checkLand(cap, args), args?.task_ref);
// Row 199 (#1697, design D9b): the git socket's publish op. checkPublish's refusal is the run's.
// The gitea token is used only under the open ticket, keyed from the agent checkPublish checked,
// and the op table returns no token. A publish then ends in recordPublish or publishRefused,
// which files any refusal once, withGitea's included.
const checkPublish = (cap, args) => refusing(cap, () => broker.checkPublish(cap, args), args?.task_ref);
const withGitea = async (ticket, fn) => credentials.useFor(broker.publishSession(ticket), 'gitea', fn);
// The push itself runs here, where the token is (D9 step 5), and the git socket's host polls
// for its end: the host's op channel answers one op at a time within 10 s, and ls-remote and
// push can take longer. Fork, user, source and dst all come from the ticket. A WorkingError's
// code is the refusal; git's stderr stays here.
const remotes = new Map();
function publishPush(ticket) {
if (remotes.has(ticket)) throw new BusError('publish-busy');
const { business, project, fork, user, source, dst } = broker.publishTarget(ticket);
const entry = { state: 'running' };
remotes.set(ticket, entry);
withGitea(ticket, (token) =>
publishRemote({ repo: workingRepo(dataRoot, business, project), fork, user, source, dst, token, tls: publishTls }).catch((e) => {
throw e instanceof WorkingError ? new BusError(e.code) : e;
}),
).then(
(r) => Object.assign(entry, { state: 'pushed', ...r }),
(e) => {
// Not a refusal: the error's name and code only, never a message that could carry git's output.
if (!(e instanceof BusError)) process.stderr.write(`bus: publish git-failed: ${JSON.stringify({ name: e?.name, code: e?.code })}\n`);
Object.assign(entry, { state: 'refused', code: e instanceof BusError ? e.code : 'git-failed' });
},
);
return { state: 'running' };
}
function publishResult(ticket) {
const entry = remotes.get(ticket);
if (!entry) throw new BusError('publish-unknown');
if (entry.state !== 'running') remotes.delete(ticket);
return { ...entry };
}
// Metadata only (service, date, state); the token stays in Credentials.
function credentialStatus(business, role) {
return credentials
@@ -286,12 +237,6 @@ export async function startBroker({
checkLand,
recordLand: (ticket, land) => broker.recordLand(ticket, land),
landRefused: (ticket, args) => broker.landRefused(ticket, args),
checkPublish,
withGitea,
publishPush,
publishResult,
recordPublish: (ticket, pushed) => broker.recordPublish(ticket, pushed),
publishRefused: (ticket, args) => broker.publishRefused(ticket, args),
landHeld: (record) => broker.landHeld(record),
// Row 156 (#1651): the launcher's landing pass.
landHolders: (business) => broker.landHolders(business),
@@ -1,182 +0,0 @@
// Row 199 (#1697, design D1, D3a, D4, section 6, T1 and T5): the per-agent credential store under
// <configDir>/secrets/agents/<business>/<agent>/<service>. Fake tokens in a scratch root only.
import test from 'node:test';
import assert from 'node:assert/strict';
import { chmodSync, existsSync, mkdirSync, mkdtempSync, readdirSync, readFileSync, rmSync, statSync, symlinkSync, writeFileSync } from 'node:fs';
import { tmpdir } from 'node:os';
import { join } from 'node:path';
import { Credentials } from '../src/credentials.mjs';
const TOKEN = 'fake_store_token_0123456789';
const ROTATED = 'fake_rotated_token_9876543210';
const AGENT = { business: 'acme', role: 'coder' };
function setup(t) {
const root = mkdtempSync(join(tmpdir(), 'bus-store-'));
t.after(() => rmSync(root, { recursive: true, force: true }));
const storeRoot = join(root, 'secrets', 'agents');
mkdirSync(storeRoot, { recursive: true, mode: 0o700 });
chmodSync(join(root, 'secrets'), 0o700);
const file = join(storeRoot, 'acme', 'coder', 'gitea');
const references = { 'acme/coder': { gitea: { service: 'gitea', store: true, file, rotateBy: '2099-01-01' } } };
const open = (extra = {}) => new Credentials({ references, storeRoot, repoRoots: [], dataRoot: '/nonexistent-data', ...extra });
const state = (c) => c.status().find((e) => e.instance === 'acme/coder' && e.service === 'gitea').state;
const read = (c, session = AGENT) => c.useFor(session, 'gitea', async (token) => ({ same: token === TOKEN, rotated: token === ROTATED }));
return { root, storeRoot, file, references, open, state, read };
}
test('T1: a store reference with no file yet starts, and the agent is not-issued until issue', async (t) => {
const s = setup(t);
const c = s.open();
assert.equal(s.state(c), 'not-issued');
await assert.rejects(s.read(c), /credential-unavailable/);
const status = c.issue('acme/coder', 'gitea', TOKEN, '2099-01-01');
assert.equal(status.state, 'valid');
assert.deepEqual(await s.read(c), { same: true, rotated: false });
assert.equal(statSync(s.file).mode & 0o777, 0o600);
for (const dir of [join(s.storeRoot, 'acme'), join(s.storeRoot, 'acme', 'coder')]) assert.equal(statSync(dir).mode & 0o777, 0o700);
assert.deepEqual(readdirSync(join(s.storeRoot, 'acme', 'coder')), ['gitea']);
// A restart reads the issued file.
assert.deepEqual(await s.read(s.open()), { same: true, rotated: false });
});
test('T1: a store path the reference does not own, or no store root, refuses the reference', (t) => {
const s = setup(t);
const other = join(s.storeRoot, 'acme', 'pm', 'gitea');
assert.throws(() => s.open({ references: { 'acme/coder': { gitea: { store: true, file: other, rotateBy: '2099-01-01' } } } }), /credential-reference/);
assert.throws(() => new Credentials({ references: s.references, repoRoots: [], dataRoot: '/nonexistent-data' }), /credential-reference/);
assert.throws(() => s.open({ references: { 'acme/@tester': { gitea: { store: true, file: join(s.storeRoot, 'acme', '@tester', 'gitea'), rotateBy: '2099-01-01' } } } }),
/credential-reference/);
});
test('T1: a wrong mode on the store root, a business dir, an agent dir or the file refuses start', (t) => {
const s = setup(t);
s.open().issue('acme/coder', 'gitea', TOKEN, '2099-01-01');
for (const [path, bad, good] of [
[s.storeRoot, 0o750, 0o700],
[join(s.storeRoot, 'acme'), 0o705, 0o700],
[join(s.storeRoot, 'acme', 'coder'), 0o770, 0o700],
[s.file, 0o640, 0o600],
]) {
chmodSync(path, bad);
assert.throws(() => s.open(), /credential-file/, path);
chmodSync(path, good);
}
s.open();
});
test('T1: a missing store root refuses start; a missing {file} reference still refuses start', (t) => {
const s = setup(t);
rmSync(s.storeRoot, { recursive: true });
assert.throws(() => s.open(), /credential-file/);
assert.throws(() => s.open({ references: { coder: { gitea: { file: join(s.root, 'absent'), rotateBy: '2099-01-01' } } } }), /credential-file/);
});
test('T1: a store file that is a directory or a symlink refuses start', (t) => {
const s = setup(t);
mkdirSync(join(s.storeRoot, 'acme', 'coder'), { recursive: true, mode: 0o700 });
chmodSync(join(s.storeRoot, 'acme'), 0o700);
mkdirSync(s.file, { mode: 0o700 });
assert.throws(() => s.open(), /credential-file/);
rmSync(s.file, { recursive: true });
const target = join(s.root, 'target');
writeFileSync(target, TOKEN + '\n', { mode: 0o600 });
symlinkSync(target, s.file);
assert.throws(() => s.open(), /credential-file/);
});
test('T2: useFor keys from the session only', async (t) => {
const s = setup(t);
const c = s.open();
c.issue('acme/coder', 'gitea', TOKEN, '2099-01-01');
await assert.rejects(s.read(c, { business: 'acme', role: 'pm' }), /credential-unavailable/);
await assert.rejects(s.read(c, { business: 'acme', role: '@tester' }), /credential-not-yours/);
await assert.rejects(s.read(c, null), /credential-not-yours/);
await assert.rejects(s.read(c, { business: 'acme' }), /credential-not-yours/);
});
test('T5: issue, rotate, revoke and delete are live with no restart, and revoke holds across one', async (t) => {
const s = setup(t);
const c = s.open();
let committed = 0;
c.issue('acme/coder', 'gitea', TOKEN, '2099-01-01', () => committed++);
assert.equal(committed, 1);
// Rotate is a second issue: the new token is used at once, with no tmp file left.
c.issue('acme/coder', 'gitea', ROTATED + '\n', '2099-06-01');
assert.deepEqual(await s.read(c), { same: false, rotated: true });
assert.equal(c.status()[0].date, '2099-06-01');
assert.deepEqual(readdirSync(join(s.storeRoot, 'acme', 'coder')), ['gitea']);
// Revoke: memory first, then the file.
assert.deepEqual(c.revoke('acme/coder', 'gitea'), { file: true });
assert.equal(s.state(c), 'revoked');
assert.equal(existsSync(s.file), false);
await assert.rejects(s.read(c), /credential-unavailable/);
const restarted = s.open();
await assert.rejects(s.read(restarted), /credential-unavailable/);
assert.equal(s.state(restarted), 'not-issued');
// Delete removes the agent's dir.
c.issue('acme/coder', 'gitea', TOKEN, '2099-01-01');
assert.deepEqual(c.remove('acme/coder'), { services: ['gitea'] });
assert.equal(existsSync(join(s.storeRoot, 'acme', 'coder')), false);
await assert.rejects(s.read(c), /credential-unavailable/);
});
test('T5: a failed commit leaves the token on disk but not in memory, and a bad token or date writes nothing', async (t) => {
const s = setup(t);
const c = s.open();
assert.throws(() => c.issue('acme/coder', 'gitea', 'short', '2099-01-01'), /credential-format/);
assert.throws(() => c.issue('acme/coder', 'gitea', TOKEN, '2099-13-01'), /credential-date/);
assert.throws(() => c.issue('acme/pm', 'gitea', TOKEN, '2099-01-01'), /credential-unknown/);
assert.equal(existsSync(s.file), false);
assert.throws(() => c.issue('acme/coder', 'gitea', TOKEN, '2099-01-01', () => { throw new Error('commit failed'); }), /commit failed/);
assert.equal(existsSync(s.file), true);
assert.equal(s.state(c), 'not-issued');
await assert.rejects(s.read(c), /credential-unavailable/);
});
test('T5: revoke of a {file} reference forgets the token and never unlinks the file', async (t) => {
const s = setup(t);
const path = join(s.root, 'outside-token');
writeFileSync(path, TOKEN + '\n', { mode: 0o600 });
const c = s.open({ references: { 'acme/coder': { gitea: { file: path, rotateBy: '2099-01-01' } } } });
assert.deepEqual(await s.read(c), { same: true, rotated: false });
assert.deepEqual(c.revoke('acme/coder', 'gitea'), { file: false });
assert.equal(existsSync(path), true);
await assert.rejects(s.read(c), /credential-unavailable/);
});
test('T5: delete refuses an agent dir holding anything but tokens', (t) => {
const s = setup(t);
const c = s.open();
c.issue('acme/coder', 'gitea', TOKEN, '2099-01-01');
writeFileSync(join(s.storeRoot, 'acme', 'coder', 'notes'), 'x', { mode: 0o600 });
assert.throws(() => c.remove('acme/coder'), /credential-file/);
assert.equal(existsSync(join(s.storeRoot, 'acme', 'coder', 'notes')), true);
});
// Design D5: the name-keyed use() is the host's own, for @sync and the tracker's start check of each
// role's token. Every other credential use goes through useFor, keyed from the broker's session. A
// new use( call site outside this list fails here.
test('D5: every credentials use( call site is on the allowlist', () => {
const packages = new URL('../../', import.meta.url);
const allowed = new Map([
['bus/src/credentials.mjs', 1], // useFor's own call
['tasks/src/adapter.mjs', 1], // the host's use(instance) for @sync and probe
]);
const found = new Map();
const walk = (dir, rel) => {
for (const name of readdirSync(dir)) {
if (name === 'node_modules' || name === 'tests') continue;
const path = join(dir, name);
if (statSync(path).isDirectory()) walk(path, `${rel}${name}/`);
else if (name.endsWith('.mjs') || name.endsWith('.js')) {
const code = readFileSync(path, 'utf8').split('\n').filter((line) => !/^\s*\/\//.test(line)).join('\n');
const n = (code.match(/(?:credentials|this)\.use\(/g) ?? []).length;
if (n) found.set(`${rel}${name}`, n);
}
}
};
for (const pkg of readdirSync(packages)) if (statSync(new URL(`${pkg}/src`, packages), { throwIfNoEntry: false })?.isDirectory())
walk(new URL(`${pkg}/src`, packages).pathname, `${pkg}/src/`);
assert.deepEqual(Object.fromEntries([...found].sort()), Object.fromEntries([...allowed].sort()));
});
@@ -1,163 +0,0 @@
// Row 199 (#1697, design section 6, T4, T5): credentials.issue, credentials.revoke and
// credentials.delete through the runtime. Human-only; the tester revokes and never issues. Each verb
// is live with no restart and leaves one receipt in changes.jsonl with no value in it. Fake tokens in a
// scratch root only.
import test from 'node:test';
import assert from 'node:assert/strict';
import { existsSync, mkdirSync, mkdtempSync, readdirSync, readFileSync, rmSync, statSync } from 'node:fs';
import { tmpdir } from 'node:os';
import { join } from 'node:path';
import { startBroker } from '../src/runtime.mjs';
import { Broker, CREDENTIAL_VERBS } from '../src/broker.mjs';
import { Store } from '../src/store.mjs';
import { businessDoc, rolesCopy, systemFor, writeJson } from '../../business/tests/helpers.mjs';
const TOKEN = 'fake_issued_token_0123456789';
const ROTATED = 'fake_rotated_token_9876543210';
const OTHER = 'fake_other_token_5555555555';
const DATE = '2099-01-01';
const role = { authority: { withinRole: ['message.send'], crossRole: [] } };
async function setup(t) {
const root = mkdtempSync(join(tmpdir(), 'bus-cred-verbs-'));
t.after(() => rmSync(root, { recursive: true, force: true }));
const rolesDir = rolesCopy(root);
const doc = { ...businessDoc(root, 'demo'), plugins: { harness: ['pi'] } };
doc.roles.coder.credentials.gitea = { store: true, rotateBy: DATE };
writeJson(join(root, 'businesses/demo.json'), doc);
mkdirSync(join(root, 'data'), { mode: 0o700 });
mkdirSync(join(root, 'secrets', 'agents'), { mode: 0o700 });
const store = (agent, service) => join(root, 'secrets', 'agents', 'demo', agent, service);
// The bus's resolved business carries the same references; a store one names its derived file.
const demo = {
id: 'demo', human: 'jason', arbiters: { technical: 'coder', delivery: 'coder' }, vars: { 'bus.tester': true },
roles: {
coder: { ...role, credentials: { gitea: { store: true, file: store('coder', 'gitea'), rotateBy: DATE }, vikunja: doc.roles.coder.credentials.vikunja } },
reviewer: { ...role, credentials: doc.roles.reviewer.credentials },
},
};
const bus = await startBroker({ dataRoot: join(root, 'data'), businesses: { demo }, organization: { dir: root, rolesDir, system: systemFor(root) } });
t.after(() => bus.close());
const human = bus.broker.bindHuman({ business: 'demo', human: 'jason', via: 'cli', outsideAgent: true });
const tester = bus.broker.bindTester({ business: 'demo' });
const ask = (cap, verb, args) => bus.broker.request(cap, { verb, args });
const file = () => JSON.parse(readFileSync(join(root, 'businesses/demo.json'), 'utf8'));
const receipts = () => {
const log = join(root, 'businesses/.history/demo/changes.jsonl');
return existsSync(log) ? readFileSync(log, 'utf8').trim().split('\n').map(JSON.parse) : [];
};
const committed = () => receipts().filter((r) => r.state === 'committed');
const status = (agent) => Object.fromEntries(bus.credentialStatus('demo', agent).map((e) => [e.service, e.state]));
return { root, bus, human, tester, ask, file, receipts, committed, status, store };
}
test('row 199: the lifecycle verbs refuse every session but the human, and the tester revokes only', (t) => {
const root = mkdtempSync(join(tmpdir(), 'bus-cred-gate-'));
const s = new Store(root);
t.after(() => { s.close(); rmSync(root, { recursive: true, force: true }); });
const calls = [];
const b = new Broker({
store: s,
businesses: { demo: { id: 'demo', human: 'jason', arbiters: { technical: 'coder', delivery: 'coder' }, vars: { 'bus.tester': true }, roles: { coder: role } } },
credentialVerbs: (id, verb, args, actor) => (calls.push([verb, actor]), { ok: true }),
});
const human = b.bindHuman({ business: 'demo', human: 'jason', via: 'cli', outsideAgent: true });
const tester = b.bindTester({ business: 'demo' });
const reader = b.bindReader({ business: 'demo' });
const coder = b.bindLaunch({ business: 'demo', role: 'coder', run: 'coder-run', harness: 'pi' });
const args = {
'credentials.issue': { agent: 'coder', service: 'gitea', token: TOKEN, date: DATE },
'credentials.revoke': { agent: 'coder', reason: 'fired' },
'credentials.delete': { agent: 'coder', reason: 'fired' },
};
assert.deepEqual(CREDENTIAL_VERBS, Object.keys(args));
for (const verb of CREDENTIAL_VERBS) {
assert.throws(() => b.request(reader, { verb, args: args[verb] }), /human-required/, verb);
assert.throws(() => b.request(coder, { verb, args: args[verb] }), /human-required/, verb);
}
assert.throws(() => b.request(tester, { verb: 'credentials.issue', args: args['credentials.issue'] }), /tester-refused/);
assert.throws(() => b.request(tester, { verb: 'credentials.delete', args: args['credentials.delete'] }), /tester-refused/);
// Keys: a token on revoke, a missing reason, an unknown service, an unknown agent, an @ agent.
assert.throws(() => b.request(human, { verb: 'credentials.revoke', args: { ...args['credentials.revoke'], token: TOKEN } }), /invalid-request/);
assert.throws(() => b.request(human, { verb: 'credentials.revoke', args: { agent: 'coder' } }), /invalid-request/);
assert.throws(() => b.request(human, { verb: 'credentials.issue', args: { ...args['credentials.issue'], service: 'model' } }), /invalid-request/);
assert.throws(() => b.request(human, { verb: 'credentials.issue', args: { ...args['credentials.issue'], file: '/x' } }), /invalid-request/);
assert.throws(() => b.request(human, { verb: 'credentials.issue', args: { ...args['credentials.issue'], agent: 'pm' } }), /credential-unknown/);
assert.throws(() => b.request(human, { verb: 'credentials.revoke', args: { agent: '@sync', reason: 'x' } }), /credential-unknown/);
assert.equal(calls.length, 0);
// A refusal records its code alone, never the token.
const refused = s.all("SELECT body FROM events WHERE kind='action.refused'");
assert.ok(refused.length >= 10);
for (const { body } of refused) assert.equal(body.includes(TOKEN), false);
// delete waits for the agent's claim to end.
b.request(coder, { verb: 'role.claim' });
assert.throws(() => b.request(human, { verb: 'credentials.delete', args: args['credentials.delete'] }), /agent-running/);
b.request(coder, { verb: 'role.release' });
for (const verb of CREDENTIAL_VERBS) b.request(human, { verb, args: args[verb] });
b.request(tester, { verb: 'credentials.revoke', args: args['credentials.revoke'] });
assert.deepEqual(calls, [['credentials.issue', 'jason'], ['credentials.revoke', 'jason'], ['credentials.delete', 'jason'], ['credentials.revoke', 'tester']]);
// No lifecycle without the runtime's store.
const bare = new Broker({ store: s, businesses: { demo: { id: 'demo', human: 'jason', arbiters: { technical: 'coder', delivery: 'coder' }, roles: { coder: role } } } });
const h2 = bare.bindHuman({ business: 'demo', human: 'jason', via: 'cli', outsideAgent: true });
assert.throws(() => bare.request(h2, { verb: 'credentials.revoke', args: args['credentials.revoke'] }), /not-configured/);
});
test('T5: issue, rotate, revoke and delete are live, rewrite the business file and leave one receipt each, with no value anywhere', async (t) => {
const c = await setup(t);
const coder = { business: 'demo', role: 'coder' };
const read = (service) => c.bus.credentials.useFor(coder, service, async (token) => token === TOKEN ? 'issued' : token === ROTATED ? 'rotated' : 'other');
assert.deepEqual(c.status('coder'), { gitea: 'not-issued', vikunja: 'valid' });
// issue: the file, then the business file's reference, then memory.
const issued = c.ask(c.human, 'credentials.issue', { agent: 'coder', service: 'gitea', token: TOKEN + '\n', date: DATE });
assert.deepEqual(issued, { agent: 'coder', verb: 'issue', service: 'gitea', date: DATE, state: 'valid' });
assert.equal(await read('gitea'), 'issued');
assert.equal(statSync(c.store('coder', 'gitea')).mode & 0o777, 0o600);
assert.deepEqual(c.file().roles.coder.credentials.gitea, { store: true, rotateBy: DATE });
let last = c.committed().at(-1);
assert.equal(last.actor, 'jason');
assert.deepEqual(last.detail, { verb: 'issue', agent: 'coder', kind: ['gitea'], date: DATE });
// The same token again is one the bus holds: the secret check refuses it before any write.
const count = c.receipts().length;
assert.throws(() => c.ask(c.human, 'credentials.issue', { agent: 'coder', service: 'gitea', token: TOKEN, date: DATE }), /credential-leak/);
assert.throws(() => c.ask(c.human, 'credentials.issue', { agent: 'coder', service: 'gitea', token: 'short', date: DATE }), /credential-format/);
assert.throws(() => c.ask(c.human, 'credentials.issue', { agent: 'coder', service: 'gitea', token: OTHER, date: '2099-02-30' }), /credential-date/);
assert.equal(c.receipts().length, count);
// rotate: an issue over an issued entry, used at once.
const rotated = c.ask(c.human, 'credentials.issue', { agent: 'coder', service: 'gitea', token: ROTATED, date: '2099-06-01' });
assert.equal(rotated.verb, 'rotate');
assert.equal(await read('gitea'), 'rotated');
assert.deepEqual(c.file().roles.coder.credentials.gitea, { store: true, rotateBy: '2099-06-01' });
assert.deepEqual(c.committed().at(-1).detail, { verb: 'rotate', agent: 'coder', kind: ['gitea'], date: '2099-06-01' });
// revoke of a store entry by the tester: memory and file, no document change, a receipt.
const bytes = readFileSync(join(c.root, 'businesses/demo.json'), 'utf8');
const revoked = c.ask(c.tester, 'credentials.revoke', { agent: 'coder', service: 'gitea', reason: 'test' });
assert.deepEqual(revoked, { agent: 'coder', verb: 'revoke', services: [{ service: 'gitea', state: 'revoked' }] });
await assert.rejects(read('gitea'), /credential-unavailable/);
assert.equal(existsSync(c.store('coder', 'gitea')), false);
assert.equal(readFileSync(join(c.root, 'businesses/demo.json'), 'utf8'), bytes);
last = c.committed().at(-1);
assert.equal(last.actor, 'tester');
assert.deepEqual(last.changes, []);
assert.deepEqual(last.detail, { verb: 'revoke', agent: 'coder', kind: ['gitea'], reason: 'test' });
// revoke of a {file} reference: memory, the outside file kept, the reference becomes store.
const outside = c.file().roles.coder.credentials.vikunja.file;
c.ask(c.human, 'credentials.revoke', { agent: 'coder', service: 'vikunja', reason: 'moved' });
await assert.rejects(read('vikunja'), /credential-unavailable/);
assert.equal(existsSync(outside), true);
assert.deepEqual(c.file().roles.coder.credentials.vikunja, { store: true, expires: DATE });
assert.ok(c.committed().at(-1).changes.some((e) => e.path === 'roles.coder.credentials.vikunja.store' && e.new === true));
// delete: every kind revoked, the agent's dir gone, both references store.
c.ask(c.human, 'credentials.issue', { agent: 'reviewer', service: 'gitea', token: OTHER, date: DATE });
assert.equal(existsSync(c.store('reviewer', 'gitea')), true);
const deleted = c.ask(c.human, 'credentials.delete', { agent: 'reviewer', reason: 'fired' });
assert.deepEqual(deleted.services.map((e) => e.state), ['revoked', 'revoked']);
assert.equal(existsSync(join(c.root, 'secrets', 'agents', 'demo', 'reviewer')), false);
assert.deepEqual(c.file().roles.reviewer.credentials, { gitea: { store: true, rotateBy: DATE }, vikunja: { store: true, expires: DATE } });
assert.deepEqual(c.committed().at(-1).detail, { verb: 'delete', agent: 'reviewer', kind: ['gitea', 'vikunja'], reason: 'fired' });
// The status read never carries a value.
assert.deepEqual(c.ask(c.human, 'credentials', {}).map((e) => e.state).sort(), ['revoked', 'revoked', 'revoked', 'revoked']);
// T4: no token in any file under the root (receipts, business file, events, logs).
const walk = (dir) => readdirSync(dir, { withFileTypes: true }).flatMap((e) => e.isDirectory() ? walk(join(dir, e.name)) : e.isFile() ? [join(dir, e.name)] : []);
for (const path of walk(c.root))
for (const token of [TOKEN, ROTATED, OTHER]) assert.equal(readFileSync(path).includes(token), false, path);
});
+3 -7
View File
@@ -10,7 +10,7 @@ import { fixture } from './spec-fixtures.mjs';
// PRD 1.5 REQ-WAIT-1, REQ-WAIT-2 and REQ-DEC-2 (row 139, #1633).
const WAITS = ['spend', 'data.delete', 'network.expose', 'account.create', 'credential.mint', 'message.external',
'git.push.protected', 'git.push.remote', 'deploy'];
'git.push.protected', 'deploy'];
const LEFT = ['git.merge.protected', 'policy.change', 'prd.approve', 'role.revoke'];
const cite = f => ({ project: 'stack', spec_path: 'spec/north-star.md', spec_commit: f.commit });
const routed = (f, d) => ({ ...f.store.get('SELECT class, route_to FROM decisions WHERE id=?', d.id) });
@@ -36,14 +36,10 @@ test('REQ-WAIT-2: a business granting a hard wait at either class is refused; th
});
test('REQ-DEC-2: every hard wait goes to the human, cited or not, and the arbiter cannot resolve it', async t => {
const f = await fixture(t, { roles: { coder: { definition: 'coder',
authority: { withinRole: ['message.send', 'record.write', 'git.push.working'], crossRole: [] } } } });
// Row 199: a git.push.remote raise binds the task's latest push, so the task has one.
f.broker.recordPush(f.coder, { task_ref: 'vikunja:1/2', project: 'stack', branch: 'feat', ref: 'work/vikunja-stack-2',
commit: 'a'.repeat(40), previous: null });
const f = await fixture(t);
for (const action of WAITS) {
for (const args of [{}, cite(f), { ...cite(f), domain: 'technical' }]) {
const d = await f.raise({ action, ...(action === 'git.push.remote' ? { target: 'vikunja:1/2' } : {}), ...args });
const d = await f.raise({ action, ...args });
assert.deepEqual(routed(f, d), { class: 'gated', route_to: 'human' }, `${action} ${JSON.stringify(args)}`);
}
}
+2 -2
View File
@@ -22,8 +22,8 @@ test('Broker keeps the sandbox client import closure: only the listed imports, n
"import { MERGE, LandingRefusal, readLanding } from './landing.mjs';",
// Row 141 (#1635): the business ID_PATTERN for instance names. The vocabulary is already in
// RUNNER_FILES and imports nothing. Row 155 (#1649) reads THREAD_DEFINITION and END_PROTECTED
// from it too. Row 199 (#1697) reads the D9c fork parse, forkProblem and forkUrl, from it.
"import { ID_PATTERN, THREAD_DEFINITION, END_PROTECTED, forkProblem, forkUrl } from '../../business/src/vocabulary.mjs';",
// from it too.
"import { ID_PATTERN, THREAD_DEFINITION, END_PROTECTED } from '../../business/src/vocabulary.mjs';",
]);
assert.doesNotMatch(source, /\bimport\s*\(/);
assert.doesNotMatch(readFileSync(new URL('../src/landing.mjs', import.meta.url), 'utf8'), /\bimport\b/);
-274
View File
@@ -1,274 +0,0 @@
// Row 199 (#1697, design D9 steps 1 to 3 and 6, D9b): the broker side of publish. git.push.remote
// is a hard wait; its decision binds the task's latest push (pushCommit). checkPublish builds the
// agent's fork from the project's upstream and the agent's own Gitea user; recordPublish spends the
// decision; publishRefused files the refusal.
import test from 'node:test';
import assert from 'node:assert/strict';
import { fork } from 'node:child_process';
import { once } from 'node:events';
import { mkdtempSync, rmSync } from 'node:fs';
import { tmpdir } from 'node:os';
import { join } from 'node:path';
import { Store } from '../src/store.mjs';
import { Broker } from '../src/broker.mjs';
const TASK = 'vikunja:1/2';
const A = 'a'.repeat(40);
const B = 'b'.repeat(40);
const C = 'c'.repeat(40);
const identity = (user) => ({ name: user, email: `${user}@example.com`, gitea: { url: 'https://git.example', user } });
const pusher = { withinRole: ['git.push.working', 'message.send'], crossRole: [] };
function businesses(upstream = 'https://git.example/mosaic/stack') {
return {
demo: {
id: 'demo',
human: 'jason',
arbiters: { technical: 'reviewer', delivery: 'reviewer' },
projects: { stack: { root: '/nonexistent', remote: upstream }, books: { root: '/nonexistent-books' } },
roles: {
coder: { authority: pusher, identity: identity('acme-coder'), publish: ['stack'] },
second: { authority: pusher, identity: identity('acme-second'), publish: ['stack'] },
plain: { authority: pusher },
reviewer: { authority: { withinRole: ['message.send'], crossRole: [] } },
},
},
};
}
function setup(t, upstream) {
const root = mkdtempSync(join(tmpdir(), 'bus-publish-'));
const store = new Store(root);
const b = new Broker({ store, businesses: businesses(upstream) });
t.after(() => {
store.close();
rmSync(root, { recursive: true, force: true });
});
const caps = {};
for (const role of ['coder', 'second', 'plain', 'reviewer']) {
caps[role] = b.bindLaunch({ business: 'demo', role, run: `${role}-run`, harness: 'pi' });
b.request(caps[role], { verb: 'role.claim', args: {} });
}
const human = b.bindHuman({ business: 'demo', human: 'jason', via: 'cli', outsideAgent: true });
const call = (cap, verb, args) => b.request(cap, { verb, args });
const push = (commit, previous = null, role = 'coder', project = 'stack') => {
b.checkPush(caps[role], TASK);
return b.recordPush(caps[role], { task_ref: TASK, project, branch: 'feat', ref: 'work/vikunja-stack-2', commit, previous });
};
const raise = (role = 'coder') =>
call(caps[role], 'decision.raise', {
action: 'git.push.remote', target: TASK, task_ref: TASK, question: 'Publish?', blocking: false, recommendation: 'yes',
options: [{ key: 'yes', text: 'Yes' }, { key: 'no', text: 'No' }],
});
const approve = (role = 'coder') => {
const d = raise(role);
call(human, 'decision.resolve', { id: d.id, choice: 'yes' });
return d;
};
const publish = (decision, role = 'coder', extra = {}) =>
b.checkPublish(caps[role], { task_ref: TASK, project: 'stack', branch: 'feat', ...(decision ? { decision: decision.id } : {}), ...extra });
const events = (kind) => store.all('SELECT * FROM events WHERE kind=? ORDER BY seq', kind).map((e) => ({ ...e, body: JSON.parse(e.body) }));
return { b, store, caps, human, call, push, raise, approve, publish, events };
}
test('a raise with no recorded push refuses no-push; a raise binds the latest push and shows it', (t) => {
const f = setup(t);
assert.throws(() => f.raise(), /no-push/);
f.push(A);
const d = f.raise();
assert.match(d.question, new RegExp(`commit ${A}\\)$`));
const raised = f.events('action.allowed').find((e) => e.body.operation === 'decision.raise');
assert.equal(raised.body.pushCommit, A);
assert.deepEqual({ ...f.store.get('SELECT class, route_to FROM decisions WHERE id=?', d.id) }, { class: 'gated', route_to: 'human' });
});
test('happy path: the fork is the upstream repo under the agent\'s own user, the destination is host-built', (t) => {
const f = setup(t);
f.push(A);
const d = f.approve();
const r = f.publish(d);
assert.deepEqual({ ...r, ticket: typeof r.ticket }, {
ticket: 'string', task_ref: TASK, project: 'stack', fork: 'https://git.example/acme-coder/stack',
user: 'acme-coder', email: '[email protected]', source: A, dst: 'refs/heads/acme-coder/feat',
});
const done = f.b.recordPublish(r.ticket, { ref: r.dst, commit: A, previous: null });
assert.deepEqual(done, { task_ref: TASK, fork: r.fork, ref: r.dst, commit: A, previous: null });
const allowed = f.events('action.allowed').filter((e) => e.body.action === 'git.push.remote' && e.body.operation !== 'decision.raise');
assert.equal(allowed.length, 1);
assert.equal(allowed[0].subject, TASK);
assert.equal(allowed[0].body.decision, d.id);
assert.equal(allowed[0].body.fork, r.fork);
// Spent once.
assert.throws(() => f.publish(d), /decision-consumed/);
assert.throws(() => f.b.recordPublish(r.ticket, { ref: r.dst, commit: A, previous: null }), /publish-unknown/);
});
test('a hard wait: no decision refuses, and a decision for another task refuses', (t) => {
const f = setup(t);
f.push(A);
assert.throws(() => f.publish(null), /decision-required/);
const other = f.call(f.caps.coder, 'decision.raise', {
action: 'deploy', target: 'vikunja:1/3', question: 'Ship?', blocking: false, recommendation: 'yes',
options: [{ key: 'yes', text: 'Yes' }, { key: 'no', text: 'No' }],
});
f.call(f.human, 'decision.resolve', { id: other.id, choice: 'yes' });
assert.throws(() => f.publish(other), /decision-mismatch/);
const d = f.raise();
assert.throws(() => f.publish(d), /decision-not-approved/);
});
test('r4 B4: approve while the latest push is A, push B, publish refuses decision-mismatch', (t) => {
const f = setup(t);
f.push(A);
const d = f.approve();
f.push(B, A);
assert.throws(() => f.publish(d), /decision-mismatch/);
// And a push between checkPublish and recordPublish refuses the record.
const f2 = setup(t);
f2.push(A);
const d2 = f2.approve();
const r = f2.publish(d2);
f2.push(C, A);
assert.throws(() => f2.b.recordPublish(r.ticket, { ref: r.dst, commit: A, previous: null }), /decision-mismatch/);
assert.throws(() => f2.b.recordPublish(r.ticket, { ref: r.dst, commit: C, previous: null }), /decision-mismatch/);
});
test('grant and fork refusals', (t) => {
const f = setup(t);
f.push(A, null, 'plain');
// plain has no identity and no grant; its raise still binds the push, the publish refuses.
const d = f.approve('plain');
assert.throws(() => f.publish(d, 'plain'), /remote-not-granted/);
assert.throws(() => f.publish(d, 'plain', { project: 'books' }), /remote-not-granted|decision-mismatch/);
// work-not-yours: second publishes a push coder made.
const g = setup(t);
g.push(A);
const d2 = g.approve('second');
assert.throws(() => g.publish(d2, 'second'), /work-not-yours/);
});
test('remote-not-fork, compared without case, and remote-host-mismatch, at push', (t) => {
for (const [upstream, code] of [
['https://git.example/acme-coder/stack', /remote-not-fork/],
['https://git.example/ACME-Coder/stack', /remote-not-fork/],
['https://other.example/mosaic/stack', /remote-host-mismatch/],
['https://git.example:3000/mosaic/stack', /remote-host-mismatch/],
]) {
const f = setup(t, upstream);
f.push(A);
const d = f.approve();
assert.throws(() => f.publish(d), code, upstream);
}
});
test('two agents granted one project each get their own fork', (t) => {
const f = setup(t);
f.push(A);
const r1 = f.publish(f.approve());
assert.equal(r1.fork, 'https://git.example/acme-coder/stack');
const g = setup(t);
g.push(A, null, 'second');
const r2 = g.publish(g.approve('second'), 'second');
assert.equal(r2.fork, 'https://git.example/acme-second/stack');
assert.equal(r2.dst, 'refs/heads/acme-second/feat');
});
test('the request cannot name the fork, the user or the destination', (t) => {
const f = setup(t);
f.push(A);
const d = f.approve();
for (const extra of [{ fork: 'https://git.example/x/stack' }, { dst: 'refs/heads/main' }, { user: 'x' }, { remote: 'https://x' }])
assert.throws(() => f.publish(d, 'coder', extra), /invalid-request/);
});
test('publishRefused files the code on the task\'s trail and drops the ticket', (t) => {
const f = setup(t);
f.push(A);
const r = f.publish(f.approve());
assert.equal(f.b.publishRefused(r.ticket, { code: 'remote-unrelated' }), 'remote-unrelated');
const refused = f.events('action.refused').filter((e) => e.subject === TASK);
assert.deepEqual(refused.at(-1).body, { code: 'remote-unrelated' });
assert.throws(() => f.b.publishRefused(r.ticket, { code: 'x' }), /publish-unknown/);
});
test('git.push.remote is in neither ROUTED list and no shipped role grants it', async () => {
const { ACTIONS, HARD_WAITS } = await import('../src/broker.mjs');
const vocabulary = await import('../../business/src/vocabulary.mjs');
assert.ok(ACTIONS.includes('git.push.remote') && HARD_WAITS.includes('git.push.remote'));
assert.ok(!vocabulary.ROUTED.includes('git.push.remote'));
const { readdirSync, readFileSync } = await import('node:fs');
const dir = new URL('../../../roles/', import.meta.url);
for (const name of readdirSync(dir).filter((n) => n.endsWith('.json')))
assert.ok(!readFileSync(new URL(name, dir), 'utf8').includes('git.push.remote'), name);
});
// D1 and D9b through the real startBroker: the store is <configDir>/secrets/agents, and the gitea
// token is used only under an open publish ticket, keyed from the agent checkPublish checked.
test('runtime: withGitea keys from the open ticket only, and the store root is the organization dir', async (t) => {
const { startBroker } = await import('../src/runtime.mjs');
const { businessDoc, rolesCopy, systemFor, writeJson } = await import('../../business/tests/helpers.mjs');
const { chmodSync, mkdirSync } = await import('node:fs');
const root = mkdtempSync(join(tmpdir(), 'bus-publish-rt-'));
t.after(() => rmSync(root, { recursive: true, force: true }));
const rolesDir = rolesCopy(root);
// The runtime reads the harness switch from the organization file as it stands.
writeJson(join(root, 'businesses/demo.json'), { ...businessDoc(root, 'demo'), plugins: { harness: ['pi'] } });
mkdirSync(join(root, 'data'), { mode: 0o700 });
mkdirSync(join(root, 'secrets', 'agents'), { recursive: true, mode: 0o700 });
chmodSync(join(root, 'secrets'), 0o700);
const demo = businesses().demo;
demo.roles.coder.credentials = { gitea: { store: true, file: join(root, 'secrets', 'agents', 'demo', 'coder', 'gitea'), rotateBy: '2099-01-01' } };
const organization = { dir: root, rolesDir, system: systemFor(root) };
// No organization, no store root: a store reference refuses the start.
await assert.rejects(startBroker({ dataRoot: join(root, 'data'), businesses: { demo } }), /credential-reference/);
const bus = await startBroker({ dataRoot: join(root, 'data'), businesses: { demo }, organization });
t.after(() => bus.close());
assert.deepEqual(bus.credentialStatus('demo', 'coder'), [{ service: 'gitea', date: '2099-01-01', state: 'not-issued' }]);
const coder = bus.broker.bindLaunch({ business: 'demo', role: 'coder', run: 'coder-run', harness: 'pi' });
const human = bus.broker.bindHuman({ business: 'demo', human: 'jason', via: 'cli', outsideAgent: true });
bus.broker.request(coder, { verb: 'role.claim', args: {} });
bus.checkPush(coder, TASK);
bus.recordPush(coder, { task_ref: TASK, project: 'stack', branch: 'feat', ref: 'work/vikunja-stack-2', commit: A, previous: null });
const d = bus.broker.request(coder, { verb: 'decision.raise', args: {
action: 'git.push.remote', target: TASK, task_ref: TASK, question: 'Publish?', blocking: false, recommendation: 'yes',
options: [{ key: 'yes', text: 'Yes' }, { key: 'no', text: 'No' }] } });
bus.broker.request(human, { verb: 'decision.resolve', args: { id: d.id, choice: 'yes' } });
assert.throws(() => bus.checkPublish(coder, { task_ref: TASK, project: 'stack', branch: 'feat' }), /decision-required/);
const ticket = bus.checkPublish(coder, { task_ref: TASK, project: 'stack', branch: 'feat', decision: d.id }).ticket;
await assert.rejects(bus.withGitea(ticket, async () => 1), /credential-unavailable/);
bus.credentials.issue('demo/coder', 'gitea', 'fake_publish_token_0123456789', '2099-01-01');
assert.equal(await bus.withGitea(ticket, async (token) => token === 'fake_publish_token_0123456789'), true);
await assert.rejects(bus.withGitea(ticket, async (token) => ({ echo: token })), /credential-leak/);
await assert.rejects(bus.withGitea('not-a-ticket', async () => 1), /publish-unknown/);
await assert.rejects(bus.withGitea(coder, async () => 1), /publish-unknown/);
bus.recordPublish(ticket, { ref: 'refs/heads/acme-coder/feat', commit: A, previous: null });
await assert.rejects(bus.withGitea(ticket, async () => 1), /publish-unknown/);
});
// Row 199 (D9b, deviation 11): the bus process answers the socket's five publish ops from their own
// launchOp lines. A LAUNCH_OPS name without one would get launch-op-refused. A boot config that
// carries publishTls still boots; process.mjs drops it before startBroker.
test('the bus process answers the publish ops from their own lines, not launch-op-refused', async (t) => {
const root = mkdtempSync(join(tmpdir(), 'bus-publish-ops-'));
t.after(() => rmSync(root, { recursive: true, force: true }));
const child = fork(new URL('../src/process.mjs', import.meta.url), [], { stdio: ['ignore', 'ignore', 'ignore', 'ipc'] });
t.after(() => {
if (child.exitCode === null) child.kill('SIGKILL');
});
const ask = async (m) => {
const reply = once(child, 'message');
child.send(m);
return (await reply)[0];
};
const config = { dataRoot: root, businesses: businesses(), launches: [], readers: [], publishTls: ['-c', 'http.sslCAInfo=/nonexistent'] };
assert.equal((await ask({ op: 'boot', config })).ok, true);
const unknown = 'not-a-ticket';
assert.deepEqual(await ask({ op: 'publishPush', ticket: unknown, id: 1 }), { ok: false, error: 'publish-unknown', id: 1 });
assert.deepEqual(await ask({ op: 'publishResult', ticket: unknown, id: 2 }), { ok: false, error: 'publish-unknown', id: 2 });
assert.deepEqual(await ask({ op: 'recordPublish', ticket: unknown, pushed: { ref: 'refs/heads/x/y', commit: A, previous: null }, id: 3 }), { ok: false, error: 'publish-unknown', id: 3 });
assert.deepEqual(await ask({ op: 'publishRefused', ticket: unknown, args: { code: 'x' }, id: 4 }), { ok: false, error: 'publish-unknown', id: 4 });
assert.deepEqual(await ask({ op: 'checkPublish', cap: 'not-a-cap', args: { task_ref: TASK, project: 'stack', branch: 'feat' }, id: 5 }), { ok: false, error: 'unauthenticated', id: 5 });
const exit = once(child, 'exit');
child.send({ op: 'close' });
assert.equal((await exit)[0], 0);
});
+8 -57
View File
@@ -10,7 +10,6 @@ import { checkVars } from "./vars.mjs";
import { parseCredentialRef } from "./credentials.mjs";
import { loadRole } from "./role.mjs";
import { checkPlugins } from "../../plugins/src/kinds.mjs";
import { forkProblem, giteaName, parseGiteaUrl } from "./vocabulary.mjs";
import {
requireObject, rejectUnknownKeys, requireId, requirePositiveInt, requireString, requireDistinctList, readJsonFile, deepFreeze,
} from "./util.mjs";
@@ -41,7 +40,7 @@ function checkBot(tracker, where) {
return { bot: tracker.bot, botId: tracker.botId };
}
function checkCredentialMap(map, services, where, store = null) {
function checkCredentialMap(map, services, where) {
requireObject(map, where);
const names = Object.keys(map).sort();
const wanted = [...services].sort();
@@ -49,37 +48,10 @@ function checkCredentialMap(map, services, where, store = null) {
refuse(`${where} must reference exactly the services its role definition needs (${wanted.join(", ") || "none"}; got ${names.join(", ") || "none"})`);
}
const out = {};
for (const service of names) out[service] = parseCredentialRef(map[service], service, `${where}.${service}`, { store });
for (const service of names) out[service] = parseCredentialRef(map[service], service, `${where}.${service}`);
return out;
}
// Row 199 (#1697, design D8): the agent's own git identity. The email is the one its Gitea user
// has, so Gitea links the commits to that user; gitea.url is the instance its pushes go to.
const EMAIL = /^[A-Za-z0-9._%+-]{1,64}@[A-Za-z0-9.-]{1,190}\.[A-Za-z]{2,24}$/;
const NAME = /^[^\0\n\r<>]{1,100}$/;
// Row 199 (design D9c): the parse is vocabulary.mjs's, where the broker reads it; this wraps it with
// the loader's refusals. Returns {url, owner, repo}.
export function httpsUrl(value, where, { repo = false } = {}) {
const parsed = parseGiteaUrl(value, { repo });
if (parsed.problem === "form") refuse(`${where} must be a normalized https URL with no user, query or fragment`);
if (parsed.problem === "instance") refuse(`${where} must name the Gitea instance only, https://HOST[:PORT]`);
if (parsed.problem === "repository") refuse(`${where} must name a repository, https://HOST/OWNER/REPO`);
return parsed;
}
function checkIdentity(identity, where) {
requireObject(identity, where);
rejectUnknownKeys(identity, ["name", "email", "gitea"], where);
if (typeof identity.name !== "string" || !NAME.test(identity.name)) refuse(`${where}.name must be 1 to 100 characters with no <, > or line break`);
if (typeof identity.email !== "string" || !EMAIL.test(identity.email)) refuse(`${where}.email must be an email address`);
requireObject(identity.gitea, `${where}.gitea`);
rejectUnknownKeys(identity.gitea, ["url", "user"], `${where}.gitea`);
httpsUrl(identity.gitea.url, `${where}.gitea.url`);
if (!giteaName(identity.gitea.user)) refuse(`${where}.gitea.user must be a Gitea user name`);
return { name: identity.name, email: identity.email, gitea: { url: identity.gitea.url, user: identity.gitea.user } };
}
function checkTracker(tracker, file) {
const where = `${file} tracker`;
requireObject(tracker, where);
@@ -102,7 +74,7 @@ function checkTracker(tracker, file) {
return { sync, labels };
}
function checkRoles(roles, file, rolesDir, business) {
function checkRoles(roles, file, rolesDir) {
const where = `${file} roles`;
requireObject(roles, where);
if (Object.keys(roles).length === 0) refuse(`${where} must declare at least one role instance`);
@@ -114,7 +86,7 @@ function checkRoles(roles, file, rolesDir, business) {
requireId(instance, `${where} instance name`);
const at = `${where}.${instance}`;
requireObject(entry, at);
rejectUnknownKeys(entry, ["definition", "holder", "vars", "tracker", "credentials", "identity", "publish"], at);
rejectUnknownKeys(entry, ["definition", "holder", "vars", "tracker", "credentials"], at);
requireId(entry.definition, `${at}.definition`);
const definition = Object.hasOwn(definitions, entry.definition) ? definitions[entry.definition] : loadRole(rolesDir, entry.definition);
definitions[entry.definition] = definition;
@@ -136,18 +108,8 @@ function checkRoles(roles, file, rolesDir, business) {
holder: entry.holder === undefined ? null : requireId(entry.holder, `${at}.holder`),
vars: checkVars(entry.vars ?? {}, "agent", `${at}.vars`),
tracker,
credentials: checkCredentialMap(entry.credentials ?? {}, services, `${at}.credentials`,
{ configDir: dirname(dirname(file)), business, agent: instance }),
credentials: checkCredentialMap(entry.credentials ?? {}, services, `${at}.credentials`),
};
// Row 199 (#1697, design D8, D9 step 2): identity and publish are optional, so today's files
// load unchanged. publish needs both an identity and a gitea credential; the project ids it
// names are checked against projects once those are read.
if (entry.identity !== undefined) out[instance].identity = checkIdentity(entry.identity, `${at}.identity`);
if (entry.publish !== undefined) {
// N10: one message for both, naming the grant.
if (!out[instance].identity || !services.includes("gitea")) refuse(`${at}.publish needs identity and a gitea definition`);
out[instance].publish = requireDistinctList(entry.publish, `${at}.publish`, (p) => requireId(p, `${at}.publish entry`), { nonEmpty: true });
}
}
return { roles: out, definitions };
}
@@ -197,7 +159,7 @@ export function validateBusinessDocument(document, file, { rolesDir }) {
}
const human = requireId(document.human, "business human");
const vars = checkVars(document.vars ?? {}, "business", `${file} vars`);
const { roles, definitions } = checkRoles(document.roles, file, rolesDir, document.id);
const { roles, definitions } = checkRoles(document.roles, file, rolesDir);
requireObject(document.arbiters, `${file} arbiters`);
rejectUnknownKeys(document.arbiters, ["delivery", "technical"], `${file} arbiters`);
@@ -221,7 +183,7 @@ export function validateBusinessDocument(document, file, { rolesDir }) {
for (const [id, entry] of Object.entries(document.projects)) {
requireId(id, `${file} project id`);
requireObject(entry, `${file} projects.${id}`);
rejectUnknownKeys(entry, ["root", "repo", "remote"], `${file} projects.${id}`);
rejectUnknownKeys(entry, ["root", "repo"], `${file} projects.${id}`);
if (typeof entry.root !== "string" || !isAbsolute(entry.root) || normalize(entry.root) !== entry.root || entry.root.includes("\0")) {
refuse(`${file} projects.${id}.root must be a normalized absolute path`);
}
@@ -229,18 +191,7 @@ export function validateBusinessDocument(document, file, { rolesDir }) {
normalize(entry.repo) !== entry.repo || entry.repo.includes("\0"))) {
refuse(`${file} projects.${id}.repo must be a normalized absolute path`);
}
// Row 199 (#1697, design D9 step 2): the upstream; a granted agent's work goes to its fork.
if (Object.hasOwn(entry, "remote")) httpsUrl(entry.remote, `${file} projects.${id}.remote`, { repo: true });
projects[id] = { root: entry.root, ...(Object.hasOwn(entry, "repo") ? { repo: entry.repo } : {}),
...(Object.hasOwn(entry, "remote") ? { remote: entry.remote } : {}) };
}
for (const [instance, role] of Object.entries(roles)) {
for (const project of role.publish ?? []) {
if (!Object.hasOwn(projects, project)) refuse(`${file} roles.${instance}.publish names ${project}, which the business file doesn't declare`);
if (!projects[project].remote) refuse(`${file} roles.${instance}.publish names ${project}, which has no "remote"`);
const problem = forkProblem(projects[project].remote, role.identity);
if (problem) refuse(`${file} roles.${instance}.publish names ${project}: ${problem}`);
}
projects[id] = { root: entry.root, ...(Object.hasOwn(entry, "repo") ? { repo: entry.repo } : {}) };
}
const tracker = checkTracker(document.tracker, file);
+8 -29
View File
@@ -3,7 +3,7 @@
// lstat and realpath only; nothing in this package opens a token file.
import { lstatSync, realpathSync } from "node:fs";
import { isAbsolute, join, normalize, relative, sep } from "node:path";
import { isAbsolute, normalize, relative, sep } from "node:path";
import { refuse } from "./errors.mjs";
import { SERVICES } from "./vocabulary.mjs";
import { requireObject, rejectUnknownKeys, requireDate } from "./util.mjs";
@@ -14,35 +14,21 @@ const WARN_DAYS = 7;
const DAY_MS = 24 * 60 * 60 * 1000;
// Shape check for one reference. Returns a frozen { service, file | env,
// rotateBy | expires }. Row 199 (#1697, design D1 and D2): `store` is the
// file at <configDir>/secrets/agents/<business>/<agent>/<service>, derived
// here and never written, so an entry can't name another agent's file. A
// reference with no agent (the tracker's sync bot) passes store = null and
// can't use the form. A `file` reference into the agents' store must be the
// agent's own path (`credential-not-yours`).
export function parseCredentialRef(ref, service, where, { store = null } = {}) {
// rotateBy | expires }.
export function parseCredentialRef(ref, service, where) {
if (!SERVICES.includes(service)) refuse(`${where}: unknown credential service ${JSON.stringify(service)}`);
requireObject(ref, where);
const dateKey = DATE_KEY[service];
rejectUnknownKeys(ref, ["file", "env", "store", dateKey], where);
rejectUnknownKeys(ref, ["file", "env", dateKey], where);
const hasFile = ref.file !== undefined;
const hasEnv = ref.env !== undefined;
const hasStore = ref.store !== undefined;
if (Number(hasFile) + Number(hasEnv) + Number(hasStore) !== 1) refuse(`${where} must name exactly one of "file", "env" or "store"`);
if (hasFile === hasEnv) refuse(`${where} must name exactly one of "file" or "env"`);
const out = { service };
if (hasStore) {
if (ref.store !== true) refuse(`${where}.store must be true`);
if (!store) refuse(`${where}: a "store" reference needs an agent`);
out.store = true;
out.file = storePath(store, service);
} else if (hasFile) {
if (hasFile) {
if (typeof ref.file !== "string" || !isAbsolute(ref.file) || normalize(ref.file) !== ref.file || ref.file.includes("\0")) {
refuse(`${where}.file must be a normalized absolute path`);
}
out.file = ref.file;
if (store && inside(join(store.configDir, "secrets", "agents"), ref.file) && ref.file !== storePath(store, service)) {
refuse(`${where}.file credential-not-yours: ${ref.file} is in the agents' store but not this agent's path`);
}
} else {
if (typeof ref.env !== "string" || !ENV_NAME.test(ref.env)) refuse(`${where}.env must match ${ENV_NAME}`);
out.env = ref.env;
@@ -52,11 +38,6 @@ export function parseCredentialRef(ref, service, where, { store = null } = {}) {
return Object.freeze(out);
}
// The store path for one agent and service (design D1).
export function storePath({ configDir, business, agent }, service) {
return join(configDir, "secrets", "agents", business, agent, service);
}
function inside(root, path) {
const rel = relative(root, path);
return rel === "" || (!rel.startsWith(`..${sep}`) && rel !== ".." && !isAbsolute(rel));
@@ -74,10 +55,8 @@ export function checkCredentialRef(ref, { forbiddenRoots = [], now = new Date(),
let stat = null;
try {
stat = lstatSync(ref.file);
} catch (e) {
// Row 199 (design D3a): a store reference with no file is an agent not issued yet.
if (ref.store && e.code === "ENOENT") warnings.push(`${label}: not issued`);
else problems.push(`${label}: not found`);
} catch {
problems.push(`${label}: not found`);
}
if (stat) {
if (stat.isSymbolicLink() || !stat.isFile()) problems.push(`${label}: must be a regular file, not a symbolic link`);
+3 -22
View File
@@ -174,9 +174,7 @@ export function organizationWriter({ dir, rolesDir, barrier = () => {} }) {
}
}
// Row 199 (#1697, design section 6): `detail` says what a credential verb did (verb, agent, kind,
// date, reason), never a value; the other setters pass none.
function write(id, document, { revision, actor, detail }) {
function write(id, document, { revision, actor }) {
requireId(actor, "authenticated actor");
requireId(id, "organization id");
return withLock(id, ({ home, parent }) => {
@@ -203,7 +201,7 @@ export function organizationWriter({ dir, rolesDir, barrier = () => {} }) {
syncDir(tx);
const event = {
transaction, actor, business: id, before, after,
changes: changes(old, written), ...(detail === undefined ? {} : { detail }), at: new Date().toISOString(),
changes: changes(old, written), at: new Date().toISOString(),
};
log(id, { ...event, state: "prepared" });
syncDir(home);
@@ -221,22 +219,5 @@ export function organizationWriter({ dir, rolesDir, barrier = () => {} }) {
});
}
// Row 199: a receipt with no document change, for a credential verb that changes only the agents'
// store (a revoke or delete of store references). Committed at once; there is nothing to recover.
function record(id, { actor, detail }) {
requireId(actor, "authenticated actor");
requireId(id, "organization id");
return withLock(id, () => {
recover(id);
const hash = sha256(regular(businessFilePath(id, dir)));
const transaction = randomUUID();
log(id, {
transaction, actor, business: id, before: hash, after: hash, changes: [], detail,
at: new Date().toISOString(), state: "committed",
});
return { revision: hash, transaction };
});
}
return { logFile, read, write, record, recover: id => withLock(id, () => recover(id)) };
return { logFile, read, write, recover: id => withLock(id, () => recover(id)) };
}
+2 -52
View File
@@ -20,9 +20,6 @@ export const ACTIONS = Object.freeze([
"git.push.working",
"git.push.protected",
"git.merge.protected",
// Row 199 (#1697, mos-1027 Q-M2): the host pushes an agent's recorded work commit to the fork its
// business grants it. A hard wait under REQ-WAIT-3 until a PRD version routes it.
"git.push.remote",
"review.request",
"review.verdict",
"record.write",
@@ -86,8 +83,8 @@ export const ROUTED = Object.freeze([
// The hard waits (PRD 1.5 REQ-WAIT-1): always gated, always the human's,
// cited spec or not. A role file may not list these under withinRole or
// crossRole (REQ-WAIT-2). Today this is spend, data.delete, network.expose,
// account.create, credential.mint, message.external, git.push.protected,
// git.push.remote and deploy; a test pins the list.
// account.create, credential.mint, message.external, git.push.protected and
// deploy; a test pins the list.
export const GATED_ONLY = Object.freeze(ACTIONS.filter((action) => !ROUTED.includes(action)));
// pi's documented built-in tools; the same list scripts/mosaic-task.mjs
@@ -136,50 +133,3 @@ export const THREAD_DEFINITION = "assistant";
// the human resolved through the CLI, may end or revoke. Row 155 moves the list here, and no
// organization write may give an instance one of them.
export const END_PROTECTED = Object.freeze([THREAD_DEFINITION]);
// Row 199 (#1697, design D9c): the Gitea URL parse. It lives here because the broker reads it and
// this file imports nothing. Gitea's name rule for users and repositories: letters, digits, "_", "."
// and "-", not "." or "..".
export const giteaName = (name) =>
typeof name === "string" && /^[A-Za-z0-9_.-]{1,100}$/.test(name) && name !== "." && name !== "..";
// An https URL that is already its own normalized form, so a dot segment, an encoded dot, userinfo, a
// query, a fragment, an upper-case host or an explicit :443 each fail, and the path checked is the
// path used. With repo, the path is exactly /OWNER/REPO (one ".git" stripped); without, it is "/".
// Returns {url, owner, repo}, or {problem} naming the rule that failed: form, instance or repository.
export function parseGiteaUrl(value, { repo = false } = {}) {
let url = null;
try {
url = typeof value === "string" ? new URL(value) : null;
} catch {
url = null;
}
const normal = url && (repo ? [url.origin + url.pathname] : [url.origin, url.origin + "/"]);
if (!url || url.protocol !== "https:" || !normal.includes(value)) return { problem: "form" };
if (!repo) return url.pathname === "/" ? { url } : { problem: "instance" };
const segments = url.pathname.split("/").slice(1);
const name = segments[1]?.endsWith(".git") ? segments[1].slice(0, -4) : segments[1];
if (segments.length !== 2 || !giteaName(segments[0]) || !giteaName(name)) return { problem: "repository" };
return { url, owner: segments[0], repo: name };
}
// Design D9 step 2: the agent's fork of an upstream, https://HOST[:PORT]/<user>/REPO, or null for an
// upstream that doesn't parse. The owner is always the agent's own Gitea user; nothing else names it.
// URL.host drops a default port, so 443 counts the same on both sides.
export function forkUrl(upstream, user) {
const { url, repo } = parseGiteaUrl(upstream, { repo: true });
return url ? `https://${url.host}/${user}/${repo}` : null;
}
// The refusal code for publishing `upstream` as an agent with `identity`, or null. Gitea compares
// names without case (lower_name), so the owner check does too. An upstream or gitea.url that doesn't
// parse is remote-not-granted: the business loader refuses both, so only a hand-built business gets
// here.
export function forkProblem(upstream, identity) {
const { url, owner } = parseGiteaUrl(upstream, { repo: true });
const instance = parseGiteaUrl(identity?.gitea?.url).url;
if (!url || !instance || typeof identity.gitea.user !== "string") return "remote-not-granted";
if (url.host !== instance.host) return "remote-host-mismatch";
if (owner.toLowerCase() === identity.gitea.user.toLowerCase()) return "remote-not-fork";
return null;
}
@@ -1,126 +0,0 @@
// Row 199 (#1697): an agent's own git identity, its publish grant, the project's upstream and the
// credential store reference. Design D1, D8, D9 step 2 and D9c.
import { test } from "node:test";
import assert from "node:assert/strict";
import { join } from "node:path";
import { BusinessError, parseCredentialRef, validateBusinessDocument } from "../src/index.mjs";
import { forkProblem, forkUrl } from "../src/vocabulary.mjs";
import { businessDoc, FAR_DATE, REPO_ROLES, tmp } from "./helpers.mjs";
function refuses(fn, pattern) {
assert.throws(fn, (error) => {
assert.ok(error instanceof BusinessError, `expected BusinessError, got ${error}`);
assert.equal(error.exitCode, 2);
assert.match(error.message, pattern);
return true;
});
}
const IDENTITY = { name: "Coder", email: "[email protected]", gitea: { url: "https://git.example", user: "acme-coder" } };
// The fixture with coder granted publish on stack, whose upstream is mosaic/stack.
function check(mutate) {
const root = tmp();
const doc = businessDoc(root);
doc.projects.stack.remote = "https://git.example/mosaic/stack";
doc.roles.coder.identity = structuredClone(IDENTITY);
doc.roles.coder.publish = ["stack"];
mutate?.(doc, root);
return validateBusinessDocument(doc, join(root, "businesses", "acme.json"), { rolesDir: REPO_ROLES });
}
test("today's files load unchanged: identity, publish and remote are optional", () => {
const root = tmp();
const business = validateBusinessDocument(businessDoc(root), join(root, "businesses", "acme.json"), { rolesDir: REPO_ROLES });
assert.equal(business.roles.coder.identity, undefined);
assert.equal(business.roles.coder.publish, undefined);
assert.equal(business.projects.stack.remote, undefined);
});
test("a granted agent loads with its identity, grant and the project's upstream", () => {
const business = check();
assert.deepEqual({ ...business.roles.coder.identity, gitea: { ...business.roles.coder.identity.gitea } }, IDENTITY);
assert.deepEqual([...business.roles.coder.publish], ["stack"]);
assert.equal(business.projects.stack.remote, "https://git.example/mosaic/stack");
});
test("N10: publish needs identity and a gitea definition, and the error names the grant", () => {
refuses(() => check((d) => { delete d.roles.coder.identity; }), /roles\.coder\.publish needs identity and a gitea definition/);
refuses(() => check((d) => { d.roles.chief.identity = structuredClone(IDENTITY); d.roles.chief.publish = ["stack"]; }),
/roles\.chief\.publish needs identity and a gitea definition/);
});
test("publish names declared projects that have a remote, each once", () => {
refuses(() => check((d) => { d.roles.coder.publish = ["books"]; }), /publish names books, which the business file doesn't declare/);
refuses(() => check((d) => { delete d.projects.stack.remote; }), /publish names stack, which has no "remote"/);
refuses(() => check((d) => { d.roles.coder.publish = []; }), /publish/);
refuses(() => check((d) => { d.roles.coder.publish = ["stack", "stack"]; }), /publish/);
});
test("identity fields are checked", () => {
refuses(() => check((d) => { d.roles.coder.identity.email = "coder"; }), /identity\.email/);
refuses(() => check((d) => { d.roles.coder.identity.name = "a\nb"; }), /identity\.name/);
refuses(() => check((d) => { d.roles.coder.identity.extra = 1; }), /identity/);
refuses(() => check((d) => { d.roles.coder.identity.gitea.user = ".."; }), /gitea\.user/);
refuses(() => check((d) => { d.roles.coder.identity.gitea.user = "a/b"; }), /gitea\.user/);
});
test("D9c: the upstream must be its own normalized form, exactly /OWNER/REPO", () => {
const bad = {
"a dot segment": "https://git.example/acme-coder/../mosaic/stack",
"an encoded dot segment": "https://git.example/acme-coder/%2e%2e/mosaic/stack",
"an extra segment": "https://git.example/mosaic/stack/more",
"a missing segment": "https://git.example/stack",
"a trailing slash": "https://git.example/mosaic/stack/",
userinfo: "https://u:[email protected]/mosaic/stack",
"a query": "https://git.example/mosaic/stack?x=1",
"a fragment": "https://git.example/mosaic/stack#f",
"an explicit :443": "https://git.example:443/mosaic/stack",
"an upper-case host": "https://GIT.example/mosaic/stack",
"a backslash": "https://git.example\\mosaic\\stack",
"a name outside Gitea's rule": "https://git.example/mosaic/st%41ck",
"a dot-dot repo": "https://git.example/mosaic/...git",
http: "http://git.example/mosaic/stack",
};
for (const [what, remote] of Object.entries(bad)) {
assert.throws(() => check((d) => { d.projects.stack.remote = remote; }), BusinessError, what);
}
});
test("D9c: one .git is stripped, and the fork is built from the stripped name", () => {
assert.equal(check((d) => { d.projects.stack.remote = "https://git.example/mosaic/stack.git"; }).projects.stack.remote,
"https://git.example/mosaic/stack.git");
assert.equal(forkUrl("https://git.example/mosaic/stack.git", "acme-coder"), "https://git.example/acme-coder/stack");
assert.equal(forkUrl("https://git.example:3000/mosaic/stack", "acme-coder"), "https://git.example:3000/acme-coder/stack");
});
test("D9c: gitea.url names the instance only", () => {
for (const url of ["https://git.example/gitea", "https://git.example/gitea/", "https://git.example/?x", "https://git.example:443", "http://git.example"]) {
assert.throws(() => check((d) => { d.roles.coder.identity.gitea.url = url; }), BusinessError, url);
}
check((d) => { d.roles.coder.identity.gitea.url = "https://git.example/"; });
});
test("the upstream's host and port must be the agent's Gitea's", () => {
refuses(() => check((d) => { d.projects.stack.remote = "https://other.example/mosaic/stack"; }), /remote-host-mismatch/);
refuses(() => check((d) => { d.projects.stack.remote = "https://git.example:3000/mosaic/stack"; }), /remote-host-mismatch/);
assert.equal(forkProblem("https://git.example/mosaic/stack", { gitea: { url: "https://git.example/", user: "u" } }), null);
});
test("remote-not-fork: an upstream the agent's own user owns, compared without case", () => {
refuses(() => check((d) => { d.projects.stack.remote = "https://git.example/acme-coder/stack"; }), /remote-not-fork/);
refuses(() => check((d) => { d.projects.stack.remote = "https://git.example/ACME-Coder/stack"; }), /remote-not-fork/);
});
test("store references: own path only (design D1, D3a)", () => {
const store = { configDir: "/cfg", business: "acme", agent: "coder" };
const ref = parseCredentialRef({ store: true, rotateBy: FAR_DATE }, "gitea", "r", { store });
assert.equal(ref.file, "/cfg/secrets/agents/acme/coder/gitea");
assert.equal(ref.store, true);
refuses(() => parseCredentialRef({ store: true, rotateBy: FAR_DATE }, "gitea", "r"), /store/);
refuses(() => parseCredentialRef({ store: true, file: "/a", rotateBy: FAR_DATE }, "gitea", "r", { store }), /exactly one/);
refuses(() => parseCredentialRef({ file: "/cfg/secrets/agents/acme/pm/gitea", rotateBy: FAR_DATE }, "gitea", "r", { store }),
/credential-not-yours/);
assert.equal(parseCredentialRef({ file: "/cfg/secrets/agents/acme/coder/gitea", rotateBy: FAR_DATE }, "gitea", "r", { store }).file,
"/cfg/secrets/agents/acme/coder/gitea");
});
+1 -1
View File
@@ -9,7 +9,7 @@ import { validateRoleDocument } from "../src/index.mjs";
import { tmp } from "./helpers.mjs";
const WAITS = ["spend", "data.delete", "network.expose", "account.create", "credential.mint", "message.external",
"git.push.protected", "deploy", "git.push.remote"];
"git.push.protected", "deploy"];
const LEFT = ["git.merge.protected", "policy.change", "prd.approve", "role.revoke"];
const role = (authority) => {
const dir = tmp();
-13
View File
@@ -1009,19 +1009,6 @@ names no one; the restart itself needs Mos's go or Jason's hand. A vars or launc
prints `no restart`, because the launcher reads it at its next launch check.
The Console Settings page has the same read and write (`packages/webui`).
## credentials (row 199)
`mosaic credentials [status [AGENT]] [--json] [--tester]` lists each agent's
credential kinds with date and state, never a value. `issue AGENT gitea
--rotate-by DATE` and `issue AGENT vikunja --expires DATE` read the token on
stdin (pipe it, or paste it and end input) and send it once to the bus's
`credentials.issue`; it is never an argument. `rotate` is the same over an
entry the bus holds now, and refuses before reading stdin when there is none.
`revoke AGENT [KIND] --reason TEXT` and `delete AGENT --reason TEXT` drop the
entries; `delete` refuses `agent-running` while the agent holds a claim. With
`--tester` only `status` and `revoke` run. After a rotate or revoke, delete
the old token in Gitea or Vikunja yourself; the bus only forgets it.
## Token usage
`mosaic usage [--business ID] [--json] [--tester]` reads the bus's accounting, without
-4
View File
@@ -17,7 +17,6 @@ import { projectBusCommand, projectCommand, PROJECT_BUS_USAGE, PROJECT_USAGE } f
import { resolveThread, threadCommand, threadMembers, threadsCommand, THREAD_TALK_USAGE, THREAD_USAGE, THREADS_USAGE } from "./thread.mjs";
import { limitsCommand, LIMITS_USAGE } from "./limits.mjs";
import { pluginsCommand, PLUGINS_USAGE } from "./plugins.mjs";
import { credentialsCommand, CREDENTIALS_USAGE } from "./credentials.mjs";
import { instancesCommand, INSTANCES_USAGE } from "./instances.mjs";
import { createChannelCaller } from "../../plugins/src/channel.mjs";
import { PluginError } from "../../plugins/src/kinds.mjs";
@@ -39,7 +38,6 @@ export const USAGE = `usage:
${USAGE_USAGE}
${LIMITS_USAGE}
${PLUGINS_USAGE}
${CREDENTIALS_USAGE}
${INSTANCES_USAGE}
${PROJECT_USAGE}
${PROJECT_BUS_USAGE}
@@ -578,8 +576,6 @@ export async function main(argv, io = { env: process.env, stdin: process.stdin,
if (verb === "limits") return limitsCommand(rest,io,p=>callFor(p,io,deps));
if (verb === "usage") return usageCommand(rest, io, p => callFor(p, io, deps));
if (verb === "plugins") return pluginsCommand(rest, io, (p) => callFor(p, io, deps));
// Row 199 (#1697): issue and rotate read the token on stdin.
if (verb === "credentials") return credentialsCommand(rest, io, (p) => callFor(p, io, deps));
// Row 165 (#1662): scripts/mosaic sends only `business instances` and `business instance` here.
if (verb === "business") return instancesCommand(rest, io, (p) => callFor(p, io, deps));
if (verb === "talk") return talk(rest, io, deps);
-83
View File
@@ -1,83 +0,0 @@
// `mosaic credentials` (row 199, #1697, design section 6): an agent's per-agent credentials through
// the broker's credentials, credentials.issue, credentials.revoke and credentials.delete. Every verb
// is the human's; with --tester (Q173) only status and revoke. issue and rotate read the token on
// stdin, never argv, and nothing here prints it: the broker's answer carries the kind, date and state.
import { CliError } from "./errors.mjs";
export const CREDENTIALS_USAGE = "mosaic credentials [status [AGENT]] [--json] [--tester] | issue|rotate AGENT gitea --rotate-by DATE | " +
"issue|rotate AGENT vikunja --expires DATE | revoke AGENT [KIND] --reason TEXT [--tester] | delete AGENT --reason TEXT [--business ID]";
const DATE_FLAG = Object.freeze({ gitea: "--rotate-by", vikunja: "--expires" });
const MAX_TOKEN = 16384;
function parse(args) {
const out = { positional: [], flags: new Set(), values: {} };
for (let i = 0; i < args.length; i++) {
const arg = args[i];
if (arg === "--json" || arg === "--tester") out.flags.add(arg);
else if (["--business", "--rotate-by", "--expires", "--reason"].includes(arg)) {
if (!args[i + 1] || Object.hasOwn(out.values, arg)) throw new CliError(CREDENTIALS_USAGE, 4);
out.values[arg] = args[++i];
} else if (arg.startsWith("--")) throw new CliError(CREDENTIALS_USAGE, 4);
else out.positional.push(arg);
}
return out;
}
// The whole of stdin, one token; a terminal shows what is typed, so a pipe or a paste then EOF.
async function readToken(stdin) {
let text = "";
for await (const chunk of stdin) {
text += chunk;
if (text.length > MAX_TOKEN + 2) throw new CliError("refused: the token on stdin is too long", 2);
}
if (!text.trim()) throw new CliError("refused: no token on stdin; pipe or paste it, then end input", 4);
return text;
}
// The states the bus's status() gives a held token: gitea is valid or rotation-due, vikunja valid,
// expiring or expired. not-issued and revoked hold none.
const ISSUED = new Set(["valid", "rotation-due", "expiring", "expired"]);
export async function credentialsCommand(args, io, callFor) {
const parsed = parse(args);
const [verb = "status", agent, kind, ...extra] = parsed.positional;
const v = parsed.values;
const shape = {
status: () => kind === undefined && !v["--rotate-by"] && !v["--expires"] && !v["--reason"],
issue: () => agent && DATE_FLAG[kind] && v[DATE_FLAG[kind]] && !v[kind === "gitea" ? "--expires" : "--rotate-by"] && !v["--reason"],
revoke: () => agent && (kind === undefined || DATE_FLAG[kind]) && v["--reason"] && !v["--rotate-by"] && !v["--expires"],
delete: () => agent && kind === undefined && v["--reason"] && !v["--rotate-by"] && !v["--expires"],
};
shape.rotate = shape.issue;
if (extra.length || !shape[verb]?.()) throw new CliError(CREDENTIALS_USAGE, 4);
// Q173 (duckworth N1): a token the tester issued would pass through its session.
if (parsed.flags.has("--tester") && !["status", "revoke"].includes(verb)) {
throw new CliError(`--tester takes only status and revoke\n${CREDENTIALS_USAGE}`, 4);
}
try {
const call = callFor(parsed);
let result;
if (verb === "status") {
const rows = await call("credentials", {});
result = agent ? rows.filter((r) => r.role === agent) : rows;
if (!parsed.flags.has("--json")) {
io.stdout.write(result.map((r) => `${r.role}\t${r.service}\t${r.state}\t${r.dateKind} ${r.date}`).join("\n") + (result.length ? "\n" : ""));
return;
}
} else if (verb === "issue" || verb === "rotate") {
// rotate names an entry the bus holds now; checked before stdin is read.
if (verb === "rotate") {
const held = (await call("credentials", {})).find((r) => r.role === agent && r.service === kind);
if (!ISSUED.has(held?.state)) throw new CliError(`refused: ${agent} has no issued ${kind} credential to rotate; use issue`, 2);
}
const token = await readToken(io.stdin);
result = await call("credentials.issue", { agent, service: kind, token, date: v[DATE_FLAG[kind]] });
} else if (verb === "revoke") {
result = await call("credentials.revoke", { agent, ...(kind ? { service: kind } : {}), reason: v["--reason"] });
} else result = await call("credentials.delete", { agent, reason: v["--reason"] });
io.stdout.write(JSON.stringify(result, null, 2) + "\n");
} catch (error) {
if (error instanceof CliError) throw error;
throw new CliError(error.message, 2);
}
}
+3 -72
View File
@@ -26,19 +26,6 @@
// squash that will land. The gate runs after the answer, and its
// outcome is the broker's: recordLand or landRefused, each on the
// task's trail and told to the role as a RESULT message.
// publish {"cap", "op": "publish", "task_ref", "project", "branch", "decision"}
// (row 199, #1697, design D9 to D11). The host answers {"ok": true,
// "result": {fork, ref, commit, previous}} or a refusal. The broker's
// checkPublish names the fork, the agent's email, the source commit
// and the destination; the request names none of them. publish
// (working.mjs) checks the work ref and the commits' identity under
// the repository's lock, then the bus process runs ls-remote and the
// push with the agent's gitea token (publishPush), and the host polls
// publishResult every publishPollMs: the host's op channel answers
// one op at a time within 10 s. The poll ends at publishDeadlineMs
// (publish-timeout), past git's own limit on each of the four git
// runs. recordPublish files the push; any refusal after checkPublish
// is filed once, by publishRefused.
//
// A refusal is {"ok": false, "error": "<code>"}, and the connection ends.
// The host checks, in order: identity (a role of this business), the request
@@ -77,10 +64,9 @@ import { randomBytes } from "node:crypto";
import { closeSync, createReadStream, existsSync, lstatSync, mkdirSync, openSync, rmSync, statSync, symlinkSync, writeSync, chmodSync } from "node:fs";
import { createServer } from "node:net";
import { join } from "node:path";
import { GIT_GRACE_MS, GIT_TIMEOUT_MS } from "../../seat/src/checkout.mjs";
import { gitSocketPath } from "../../seat/src/session.mjs";
import { gateEnv, isMosaicStack, rootHits, runGate } from "../../seat/src/landing.mjs";
import { BUNDLE_MAX, dropLand, finishLand, FSCK, HAVE_MAX, prepareLand, publish, push, review, validBranch, workRef, WorkingError } from "../../seat/src/working.mjs";
import { BUNDLE_MAX, dropLand, finishLand, FSCK, HAVE_MAX, prepareLand, push, review, validBranch, workRef, WorkingError } from "../../seat/src/working.mjs";
import { CliError } from "./errors.mjs";
const HEADER_MAX = 4096;
@@ -98,11 +84,6 @@ export const bodyDeadline = (size) => BODY_IDLE_MS + Math.ceil(size / BODY_MIN_R
const MAX_ACTIVE = 8;
// The longest close() waits for its host-closing answers to leave.
export const CLOSE_WAIT_MS = 2000;
// How often the host asks the bus process whether a publish's push ended.
const PUBLISH_POLL_MS = 200;
// The longest the host polls: publishRemote runs at most four git commands
// (ls-remote, cat-file, merge-base, push), each ended by git's own limit.
const PUBLISH_DEADLINE_MS = 4 * (GIT_TIMEOUT_MS + GIT_GRACE_MS) + 30_000;
const CAP = /^[0-9a-f]{64}$/;
const OID = /^[0-9a-f]{40}([0-9a-f]{24})?$/;
@@ -291,7 +272,7 @@ Applied-under: git socket land (bus host)
// host: startHost()'s handle; business: loadBusiness() output. bundleMax,
// bodyDeadline, bodyIdle, maxActive, fsck and gate (LAND_GATE's shape) are
// for the tests.
export function createGitSocket({ host, business: b, dataRoot, env = process.env, log = () => {}, bundleMax = BUNDLE_MAX, bodyDeadline: deadlineOf = bodyDeadline, bodyIdle = BODY_IDLE_MS, maxActive = MAX_ACTIVE, fsck = FSCK, gate = LAND_GATE, publishPollMs = PUBLISH_POLL_MS, publishDeadlineMs = PUBLISH_DEADLINE_MS }) {
export function createGitSocket({ host, business: b, dataRoot, env = process.env, log = () => {}, bundleMax = BUNDLE_MAX, bodyDeadline: deadlineOf = bodyDeadline, bodyIdle = BODY_IDLE_MS, maxActive = MAX_ACTIVE, fsck = FSCK, gate = LAND_GATE }) {
const path = gitSocketPath(dataRoot);
const dir = join(dataRoot, "bus-host");
let server = null;
@@ -405,55 +386,6 @@ export function createGitSocket({ host, business: b, dataRoot, env = process.env
})();
}
async function doPublish(socket, cap, req) {
const { task_ref, branch, decision } = req;
const p = project(req.project);
if (!p || !validBranch(branch) || (decision !== undefined && typeof decision !== "string")) throw new Refusal("invalid-request");
workRef(task_ref);
const facts = await op({ op: "checkPublish", cap, args: { task_ref, project: req.project, branch, ...(decision === undefined ? {} : { decision }) } });
const { ticket } = facts;
// From the ticket on, a refusal is the publish's own, filed once by publishRefused.
const refused = async (e) => {
const code = e instanceof Refusal || e instanceof Recorded || e instanceof WorkingError ? e.code : "git-failed";
if (code === "git-failed") log(`git socket: publish ${task_ref}: ${code}: ${e.message}`);
await host.op({ op: "publishRefused", ticket, args: { code } }).catch((x) => log(`git socket: publish ${task_ref}: refusal ${code} not recorded: ${x.code ?? x.message}`));
return new Recorded(code);
};
let pushed;
try {
pushed = await publish({
dataRoot, business: b.id, project: req.project, root: p.root, taskRef: task_ref, source: facts.source, email: facts.email, env,
remote: async () => {
await op({ op: "publishPush", ticket });
const until = Date.now() + publishDeadlineMs;
for (;;) {
// The bus's git may still finish after this; the fork could then hold the commit with
// no record, as when recordPublish refuses after a push.
if (Date.now() >= until) {
log(`git socket: publish ${task_ref}: publish-timeout; the bus process's push may still reach the fork`);
throw new Refusal("publish-timeout");
}
await new Promise((r) => setTimeout(r, publishPollMs));
const r = await op({ op: "publishResult", ticket });
if (r.state === "pushed") return { ref: r.ref, commit: r.commit, previous: r.previous };
if (r.state === "refused") throw new Recorded(r.code);
}
},
});
} catch (e) {
throw await refused(e);
}
// The commit is on the fork now; a record the broker refuses is filed as that refusal.
let result;
try {
result = await op({ op: "recordPublish", ticket, pushed });
} catch (e) {
log(`git socket: publish ${task_ref}: ${pushed.commit} is on ${facts.fork} but wasn't recorded: ${e.code}`);
throw await refused(e);
}
socket.end(`${JSON.stringify({ ok: true, result: { fork: result.fork, ref: result.ref, commit: result.commit, previous: result.previous } })}\n`);
}
async function serve(socket) {
let cap = null;
let subject = null;
@@ -478,8 +410,7 @@ export function createGitSocket({ host, business: b, dataRoot, env = process.env
if (req.op === "push") await doPush(socket, cap, req, rest, file);
else if (req.op === "fetch") await doFetch(socket, cap, req, file);
else if (req.op === "land") await doLand(socket, cap, req);
else if (req.op === "publish") await doPublish(socket, cap, req);
else throw new Refusal("invalid-request", "op is push, fetch, land or publish");
else throw new Refusal("invalid-request", "op is push, fetch or land");
} catch (e) {
if (e instanceof Recorded) return fail(e.code);
const code = e instanceof Refusal || e instanceof WorkingError ? e.code : "git-failed";
+3 -5
View File
@@ -47,9 +47,8 @@
// launchedBy = the business's human and no caller capability.
//
// listen() also serves the git socket (src/git-socket.mjs, row 93), which
// the sandbox binds for a role that holds git.push.working,
// review.verdict or git.push.remote (row 199), or a merge decision to land
// (landHeld, row 137);
// the sandbox binds for a role that holds git.push.working or
// review.verdict, or a merge decision to land (landHeld, row 137);
// session.json names it as gitSocket.
//
// Thread sessions (row 121, #1615). A thread instance (an instance whose
@@ -398,9 +397,8 @@ export function createLauncher({
const transcript = call(model.harness, "transcript", () => harness.transcript({ runDir, store }));
const harnessMounts = call(model.harness, "mounts", () => harness.mounts({ env }));
// Row 93: the git socket, for a role that pushes work or reviews it. Row 137: or lands it.
// Row 199 (D11): or publishes it, once a reviewed commit routes git.push.remote within-role.
const held = [...resolved.limits.authority.withinRole, ...resolved.limits.authority.crossRole];
const git = land || ["git.push.working", "review.verdict", "git.push.remote"].some((a) => held.includes(a)) ? gitSocketPath(dataRoot) : null;
const git = land || held.includes("git.push.working") || held.includes("review.verdict") ? gitSocketPath(dataRoot) : null;
writeSessionFile(runDir, {
...sessionDefaults,
business: b.id,
-114
View File
@@ -1,114 +0,0 @@
// Row 199 (#1697, design section 6, T4, T5): `mosaic credentials`. issue and rotate read the token on
// stdin and pass it once to credentials.issue; nothing printed carries it. --tester takes status and
// revoke only. A fake transport records the calls; no bus runs here.
import test from "node:test";
import assert from "node:assert/strict";
import { spawnSync } from "node:child_process";
import { join } from "node:path";
import { fileURLToPath } from "node:url";
import { main } from "../src/cli.mjs";
import { io, tmp } from "./helpers.mjs";
const TOKEN = "fake_cli_token_0123456789";
const MOSAIC = fileURLToPath(new URL("../../../scripts/mosaic", import.meta.url));
const ROWS = [
{ service: "gitea", account: null, role: "coder", dateKind: "rotateBy", date: "2099-01-01", state: "valid" },
{ service: "vikunja", account: "bot", role: "coder", dateKind: "expires", date: "2099-01-01", state: "not-issued" },
{ service: "gitea", account: null, role: "reviewer", dateKind: "rotateBy", date: "2099-01-01", state: "revoked" },
];
function deps(calls, rows = ROWS) {
const answer = (who) => (business) => async (verb, args) => {
calls.push({ who, business, verb, args });
if (verb === "credentials") return rows;
if (verb === "credentials.issue") return { agent: args.agent, verb: "issue", service: args.service, date: args.date, state: "valid" };
return { agent: args.agent, verb: verb.slice(12), services: [] };
};
return { system: { dataRoot: "/nonexistent" }, transport: answer("human"), testerTransport: answer("tester") };
}
const withStdin = (x, text) => {
x.stdin.end(text);
return x;
};
test("row 199: issue and rotate send the stdin token once to credentials.issue and print no value", async () => {
const calls = [];
const d = deps(calls);
const x = withStdin(io(), TOKEN + "\n");
await main(["credentials", "issue", "coder", "gitea", "--rotate-by", "2099-01-01", "--business", "acme"], x, d);
assert.deepEqual(calls, [{ who: "human", business: "acme", verb: "credentials.issue", args: { agent: "coder", service: "gitea", token: TOKEN + "\n", date: "2099-01-01" } }]);
assert.equal(x.out.text.includes(TOKEN), false);
assert.equal(JSON.parse(x.out.text).state, "valid");
// rotate checks the entry is issued before it reads stdin.
calls.length = 0;
await main(["credentials", "rotate", "coder", "gitea", "--rotate-by", "2099-06-01", "--business", "acme"], withStdin(io(), TOKEN), d);
assert.deepEqual(calls.map((c) => c.verb), ["credentials", "credentials.issue"]);
calls.length = 0;
await assert.rejects(main(["credentials", "rotate", "coder", "vikunja", "--expires", "2099-06-01", "--business", "acme"], withStdin(io(), TOKEN), d), /no issued vikunja credential to rotate; use issue/);
assert.deepEqual(calls.map((c) => c.verb), ["credentials"]);
// No token on stdin, the date flag of the other kind, a token in argv: refused before any write.
calls.length = 0;
await assert.rejects(main(["credentials", "issue", "coder", "gitea", "--rotate-by", "2099-01-01", "--business", "acme"], withStdin(io(), "\n"), d), /no token on stdin/);
await assert.rejects(main(["credentials", "issue", "coder", "gitea", "--expires", "2099-01-01", "--business", "acme"], withStdin(io(), TOKEN), d), /mosaic credentials/);
await assert.rejects(main(["credentials", "issue", "coder", "gitea", TOKEN, "--rotate-by", "2099-01-01", "--business", "acme"], withStdin(io(), TOKEN), d), /mosaic credentials/);
await assert.rejects(main(["credentials", "issue", "coder", "gitea", "--rotate-by", "2099-01-01", "--business", "acme"], withStdin(io(), "x".repeat(17000)), d), /too long/);
assert.equal(calls.filter((c) => c.verb === "credentials.issue").length, 0);
// From inside an agent run, the human's verbs refuse before any call.
await assert.rejects(main(["credentials", "issue", "coder", "gitea", "--rotate-by", "2099-01-01", "--business", "acme"], withStdin(io({ CLAUDECODE: "1" }), TOKEN), d), /outside any agent run/);
});
// gizmoduck-0050: a gitea key past its rotate-by date is rotation-due, and rotate is how a human
// clears it. Every state the bus's status() gives a held token rotates; the other two refuse.
test("row 199: rotate goes ahead on each held state, rotation-due included, and refuses not-issued, revoked and an unknown entry", async () => {
const states = { valid: true, "rotation-due": true, expiring: true, expired: true, "not-issued": false, revoked: false };
for (const [state, held] of Object.entries(states)) {
const calls = [];
const d = deps(calls, [{ service: "gitea", account: null, role: "coder", dateKind: "rotateBy", date: "2020-01-01", state }]);
const run = main(["credentials", "rotate", "coder", "gitea", "--rotate-by", "2099-06-01", "--business", "acme"], withStdin(io(), TOKEN), d);
if (held) await run;
else await assert.rejects(run, /no issued gitea credential to rotate; use issue/);
assert.deepEqual(calls.map((c) => c.verb), held ? ["credentials", "credentials.issue"] : ["credentials"], state);
}
const calls = [];
await assert.rejects(main(["credentials", "rotate", "tester", "gitea", "--rotate-by", "2099-06-01", "--business", "acme"], withStdin(io(), TOKEN), deps(calls)), /no issued gitea credential/);
assert.deepEqual(calls.map((c) => c.verb), ["credentials"]);
});
test("row 199: status, revoke and delete; the tester runs status and revoke only", async () => {
const calls = [];
const d = deps(calls);
const x = io();
await main(["credentials", "status", "coder", "--business", "acme"], x, d);
assert.equal(x.out.text, "coder\tgitea\tvalid\trotateBy 2099-01-01\ncoder\tvikunja\tnot-issued\texpires 2099-01-01\n");
const j = io();
await main(["credentials", "--json", "--business", "acme", "--tester"], j, d);
assert.equal(JSON.parse(j.out.text).length, 3);
await main(["credentials", "revoke", "coder", "gitea", "--reason", "fired", "--business", "acme", "--tester"], io(), d);
await main(["credentials", "revoke", "coder", "--reason", "fired", "--business", "acme"], io(), d);
await main(["credentials", "delete", "coder", "--reason", "fired", "--business", "acme"], io(), d);
assert.deepEqual(calls.map((c) => [c.who, c.verb, c.args]), [
["human", "credentials", {}],
["tester", "credentials", {}],
["tester", "credentials.revoke", { agent: "coder", service: "gitea", reason: "fired" }],
["human", "credentials.revoke", { agent: "coder", reason: "fired" }],
["human", "credentials.delete", { agent: "coder", reason: "fired" }],
]);
calls.length = 0;
for (const args of [
["issue", "coder", "gitea", "--rotate-by", "2099-01-01", "--tester"],
["rotate", "coder", "gitea", "--rotate-by", "2099-01-01", "--tester"],
["delete", "coder", "--reason", "x", "--tester"],
]) await assert.rejects(main(["credentials", ...args, "--business", "acme"], withStdin(io(), TOKEN), d), /--tester takes only status and revoke/);
for (const args of [["revoke", "coder"], ["delete", "coder"], ["delete", "coder", "gitea", "--reason", "x"], ["revoke", "coder", "model", "--reason", "x"], ["purge", "coder"]])
await assert.rejects(main(["credentials", ...args, "--business", "acme"], io(), d), /mosaic credentials/, args.join(" "));
assert.equal(calls.length, 0);
});
test("row 199: scripts/mosaic sends credentials to packages/cli, not the seat CLI", (t) => {
const root = tmp(t, "mosaic-cli-cred-route-");
const env = { PATH: process.env.PATH, HOME: join(root, "home"), MOSAIC_CONFIG: join(root, "none.json") };
const r = spawnSync(MOSAIC, ["credentials", "purge"], { env, encoding: "utf8" });
assert.equal(r.status, 4);
assert.match(r.stderr, /^mosaic: mosaic credentials \[status \[AGENT\]\]/);
assert.doesNotMatch(r.stderr, /mosaic launch </);
});
-281
View File
@@ -1,281 +0,0 @@
// Row 199 (#1697, design D9, D9b, D11, T3): the git socket's publish op end to end. The socket,
// working.mjs, the bus runtime (in this process, so its publishTls can name the scratch CA), the
// per-agent store and a TLS fake Gitea on 127.0.0.1. The host stand-in answers the socket's ops
// the way packages/bus/src/process.mjs does. Fake tokens in a scratch root only.
import { test } from "node:test";
import assert from "node:assert/strict";
import { spawnSync } from "node:child_process";
import { chmodSync, lstatSync, mkdirSync, mkdtempSync, readdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
import { connect } from "node:net";
import { tmpdir } from "node:os";
import { join } from "node:path";
import { DatabaseSync } from "node:sqlite";
import { startBroker } from "../../bus/src/runtime.mjs";
import { businessDoc, rolesCopy, systemFor, writeJson } from "../../business/tests/helpers.mjs";
import { pushWorking } from "../../harness/src/git-tools.mjs";
import { gitEnv } from "../../seat/src/checkout.mjs";
import { workRef } from "../../seat/src/working.mjs";
import { fakeGitea, scratchCa } from "../../seat/tests/fake-gitea.mjs";
import { createGitSocket } from "../src/git-socket.mjs";
const TASK = "vikunja:1/2";
const TASK2 = "vikunja:1/3";
const TOKEN = "fake_socket_token_0123456789";
const TOKEN2 = "fake_second_token_0123456789";
function git(cwd, ...args) {
const r = spawnSync("git", args, { cwd, encoding: "utf8", env: gitEnv() });
assert.equal(r.status, 0, `git ${args.join(" ")}: ${r.stderr}`);
return r.stdout.trim();
}
function commit(dir, text, email) {
writeFileSync(join(dir, "README"), `${text}\n`);
git(dir, "add", "README");
git(dir, "-c", `user.name=${text}`, "-c", `user.email=${email}`, "commit", "-q", "-m", text);
return git(dir, "rev-parse", "HEAD");
}
function raw(path, req) {
return new Promise((resolve, reject) => {
const s = connect(path);
let buf = "";
s.on("connect", () => s.write(`${JSON.stringify(req)}\n`));
s.on("data", (b) => (buf += b));
s.on("end", () => resolve(JSON.parse(buf.split("\n")[0])));
s.on("error", reject);
});
}
// The socket's ops, answered as process.mjs's launchOp answers them.
const OPS = {
identity: (r, m) => r.identity(m.cap),
refuse: (r, m) => r.refuse(m.cap, m.code, m.subject),
checkPush: (r, m) => r.checkPush(m.cap, m.task_ref),
recordPush: (r, m) => r.recordPush(m.cap, m.push),
checkPublish: (r, m) => r.checkPublish(m.cap, m.args),
publishPush: (r, m) => r.publishPush(m.ticket),
publishResult: (r, m) => r.publishResult(m.ticket),
recordPublish: (r, m) => r.recordPublish(m.ticket, m.pushed),
publishRefused: (r, m) => r.publishRefused(m.ticket, m.args),
};
async function setup(t, { publishDeadlineMs } = {}) {
const root = mkdtempSync(join(tmpdir(), "cli-publish-"));
const down = [];
t.after(async () => {
for (const fn of down) await fn();
rmSync(root, { recursive: true, force: true });
});
const ca = scratchCa(join(root, "ca"));
const gitea = await fakeGitea({ dir: join(root, "gitea"), tls: ca, users: { coder: TOKEN, second: TOKEN2 } });
down.push(() => gitea.close());
const project = join(root, "project");
mkdirSync(project);
git(project, "init", "-q", "-b", "main");
commit(project, "one", "[email protected]");
// Each agent's fork and the upstream, mosaic/stack, all clones of the root.
for (const owner of ["coder", "second", "mosaic"]) {
mkdirSync(join(gitea.repo(owner, "stack"), ".."), { recursive: true });
git(root, "clone", "-q", "--bare", project, gitea.repo(owner, "stack"));
}
const forkRepo = gitea.repo("coder", "stack");
const rolesDir = rolesCopy(root);
writeJson(join(root, "businesses/demo.json"), { ...businessDoc(root, "demo"), plugins: { harness: ["pi"] } });
mkdirSync(join(root, "data"), { mode: 0o700 });
mkdirSync(join(root, "secrets", "agents"), { recursive: true, mode: 0o700 });
chmodSync(join(root, "secrets"), 0o700);
const agent = (user) => ({
authority: { withinRole: ["git.push.working", "message.send"], crossRole: [] },
identity: { name: user, email: `${user}@example.com`, gitea: { url: gitea.origin, user } },
publish: ["stack"],
credentials: { gitea: { store: true, file: join(root, "secrets", "agents", "demo", user, "gitea"), rotateBy: "2099-01-01" } },
});
const demo = {
id: "demo", human: "jason", arbiters: { technical: "reviewer", delivery: "reviewer" },
projects: { stack: { root: project, remote: `${gitea.origin}/mosaic/stack` } },
roles: {
coder: agent("coder"),
second: agent("second"),
reviewer: { authority: { withinRole: ["message.send"], crossRole: [] } },
},
};
const dataRoot = join(root, "data");
const bus = await startBroker({ dataRoot, businesses: { demo }, organization: { dir: root, rolesDir, system: systemFor(root) }, publishTls: ["-c", `http.sslCAInfo=${ca.ca}`] });
down.unshift(() => bus.close());
// A test sets hooks[op] to act around one op: hook(message, next), where next() answers it.
const hooks = {};
const host = { op: async (m) => (hooks[m.op] ? hooks[m.op](m, () => OPS[m.op](bus, m)) : OPS[m.op](bus, m)) };
const logs = [];
mkdirSync(join(dataRoot, "bus-host"), { mode: 0o700 });
const sock = createGitSocket({ host, business: demo, dataRoot, env: process.env, log: (l) => logs.push(l), publishPollMs: 20, ...(publishDeadlineMs ? { publishDeadlineMs } : {}) });
down.unshift(() => sock.close());
await sock.listen();
const caps = {};
for (const role of ["coder", "second"]) {
caps[role] = bus.broker.bindLaunch({ business: "demo", role, run: `${role}-run`, harness: "pi" });
bus.broker.request(caps[role], { verb: "role.claim", args: {} });
}
const human = bus.broker.bindHuman({ business: "demo", human: "jason", via: "cli", outsideAgent: true });
// Each agent's workspace: a clone of the root with one commit of its own on feat, pushed to the
// working repo for its task.
const work = async (role, task) => {
const workspace = join(root, `ws-${role}`);
const tmp = join(root, `tmp-${role}`);
mkdirSync(workspace);
mkdirSync(tmp);
git(workspace, "clone", "-q", project, "stack");
const clone = join(workspace, "stack");
git(clone, "checkout", "-q", "-b", "feat");
const commitId = commit(clone, `work by ${role}`, `${role}@example.com`);
await pushWorking({ path: sock.path, cap: caps[role], workspace, tmp, task_ref: task, project: "stack", branch: "feat" });
return commitId;
};
// A second commit on the coder's feat, pushed to the working repo through the socket.
const again = async () => {
const workspace = join(root, "ws-coder");
const commitId = commit(join(workspace, "stack"), "more work", "[email protected]");
await pushWorking({ path: sock.path, cap: caps.coder, workspace, tmp: join(root, "tmp-coder"), task_ref: TASK, project: "stack", branch: "feat" });
return commitId;
};
const approve = (role = "coder", task = TASK) => {
const d = bus.broker.request(caps[role], { verb: "decision.raise", args: {
action: "git.push.remote", target: task, task_ref: task, question: "Publish?", blocking: false, recommendation: "yes",
options: [{ key: "yes", text: "Yes" }, { key: "no", text: "No" }] } });
bus.broker.request(human, { verb: "decision.resolve", args: { id: d.id, choice: "yes" } });
return d.id;
};
const publish = (decision, extra = {}, role = "coder", task = TASK) =>
raw(sock.path, { cap: caps[role], op: "publish", task_ref: task, project: "stack", branch: "feat", decision, ...extra });
// The bus holds the writer lock; the test reads its events through a read-only handle.
const events = (kind, task = TASK) => {
const db = new DatabaseSync(join(dataRoot, "bus", "bus.sqlite"), { readOnly: true });
try {
return db.prepare("SELECT * FROM events WHERE kind=? AND subject=? ORDER BY seq").all(kind, task).map((e) => ({ ...e, body: JSON.parse(e.body) }));
} finally {
db.close();
}
};
return { root, bus, gitea, forkRepo, work: await work("coder", TASK), more: work, again, hooks, caps, approve, publish, logs, events };
}
test("T3 end to end: publish under an approved decision sends the work commit to the agent's own fork and records it once", async (t) => {
const w = await setup(t);
// Nothing issued: the store has no token, and the refusal is filed once.
const first = w.approve();
assert.deepEqual(await w.publish(first), { ok: false, error: "credential-unavailable" });
assert.deepEqual(w.events("action.refused").map((e) => e.body.code), ["credential-unavailable"]);
assert.equal(w.gitea.requests.length, 0);
w.bus.credentials.issue("demo/coder", "gitea", TOKEN, "2099-01-01");
const reply = await w.publish(first);
assert.deepEqual(reply, { ok: true, result: { fork: `${w.gitea.origin}/coder/stack`, ref: "refs/heads/coder/feat", commit: w.work, previous: null } });
assert.equal(git(w.root, `--git-dir=${w.forkRepo}`, "rev-parse", "refs/heads/coder/feat"), w.work);
// Every request went to the agent's fork; the upstream, mosaic/stack, received none.
assert.ok(w.gitea.requests.some((r) => r.path === "/coder/stack/git-receive-pack" && r.user === "coder"));
for (const r of w.gitea.requests) assert.match(r.path, /^\/coder\/stack\//);
const allowed = w.events("action.allowed").filter((e) => e.body.action === "git.push.remote" && e.body.operation !== "decision.raise");
assert.equal(allowed.length, 1);
assert.equal(allowed[0].body.fork, `${w.gitea.origin}/coder/stack`);
assert.equal(allowed[0].body.commit, w.work);
// The decision is spent.
assert.deepEqual(await w.publish(first), { ok: false, error: "decision-consumed" });
// T4: no token in any answer, event or log line, nor in any file under the scratch root but the
// agent's own store file.
assert.equal(JSON.stringify([reply, w.events("action.allowed"), w.events("action.refused"), w.logs]).includes(TOKEN), false);
const store = join(w.root, "secrets", "agents", "demo", "coder", "gitea");
const holding = readdirSync(w.root, { recursive: true })
.map((name) => join(w.root, name))
.filter((file) => lstatSync(file).isFile() && readFileSync(file).includes(TOKEN));
assert.deepEqual(holding, [store]);
});
test("T3 end to end: a fork the token can't reach is fork-unavailable, filed once, and nothing is recorded as allowed", async (t) => {
const w = await setup(t);
w.bus.credentials.issue("demo/coder", "gitea", "fake_wrong_token_0123456789", "2099-01-01");
assert.deepEqual(await w.publish(w.approve()), { ok: false, error: "fork-unavailable" });
assert.ok(w.gitea.requests.length > 0);
assert.deepEqual(w.events("action.refused").map((e) => e.body.code), ["fork-unavailable"]);
assert.equal(w.events("action.allowed").filter((e) => e.body.action === "git.push.remote" && e.body.operation !== "decision.raise").length, 0);
});
test("a bad branch or project refuses before the broker; a fork, user or ref in the request is ignored", async (t) => {
const w = await setup(t);
const d = w.approve();
assert.deepEqual(await w.publish(d, { branch: "../main" }), { ok: false, error: "invalid-request" });
assert.deepEqual(await w.publish(d, { project: "nope" }), { ok: false, error: "invalid-request" });
// Filed through the socket's refuse op, as a bad push request is; no publish ticket opened.
assert.deepEqual(w.events("action.refused").map((e) => e.body.code), ["invalid-request", "invalid-request"]);
assert.equal(w.gitea.requests.length, 0);
// The broker builds the fork and the destination; the request's own are never read.
w.bus.credentials.issue("demo/coder", "gitea", TOKEN, "2099-01-01");
const reply = await w.publish(d, { fork: "https://127.0.0.1:1/other/stack", user: "other", dst: "refs/heads/main", ref: "refs/heads/main" });
assert.deepEqual(reply.result, { fork: `${w.gitea.origin}/coder/stack`, ref: "refs/heads/coder/feat", commit: w.work, previous: null });
for (const r of w.gitea.requests) assert.match(r.path, /^\/coder\/stack\//);
assert.equal(spawnSync("git", [`--git-dir=${w.forkRepo}`, "rev-parse", "-q", "--verify", "refs/heads/other/feat"], { env: gitEnv() }).status, 1);
});
test("T3 end to end: two agents, one project, each lands on its own fork only; the upstream receives nothing", async (t) => {
const w = await setup(t);
const second = await w.more("second", TASK2);
w.bus.credentials.issue("demo/coder", "gitea", TOKEN, "2099-01-01");
w.bus.credentials.issue("demo/second", "gitea", TOKEN2, "2099-01-01");
assert.equal((await w.publish(w.approve())).ok, true);
const coderRequests = w.gitea.requests.length;
for (const r of w.gitea.requests) assert.deepEqual([r.path.split("/")[1], r.user ?? "coder"], ["coder", "coder"]);
const reply = await w.publish(w.approve("second", TASK2), {}, "second", TASK2);
assert.deepEqual(reply.result, { fork: `${w.gitea.origin}/second/stack`, ref: "refs/heads/second/feat", commit: second, previous: null });
for (const r of w.gitea.requests.slice(coderRequests)) assert.deepEqual([r.path.split("/")[1], r.user ?? "second"], ["second", "second"]);
const refs = (owner) => git(w.root, `--git-dir=${w.gitea.repo(owner, "stack")}`, "for-each-ref", "--format=%(refname) %(objectname)");
assert.equal(refs("coder").includes(`refs/heads/coder/feat ${w.work}`), true);
assert.equal(refs("coder").includes("refs/heads/second/"), false);
assert.equal(refs("second").includes(`refs/heads/second/feat ${second}`), true);
assert.equal(refs("second").includes("refs/heads/coder/"), false);
assert.equal(refs("mosaic").split("\n").length, 1, "the upstream holds only main");
});
// sage-1440 on BUILD.md 14: a push between checkPublish and the git push refuses, with one code
// for each place it can come in, and the fork never receives a commit the decision didn't bind.
test("T3: a push after checkPublish and before the lock refuses ref-moved; the fork receives nothing", async (t) => {
const w = await setup(t);
w.bus.credentials.issue("demo/coder", "gitea", TOKEN, "2099-01-01");
w.hooks.checkPublish = async (m, next) => {
const facts = await next();
await w.again();
return facts;
};
assert.deepEqual(await w.publish(w.approve()), { ok: false, error: "ref-moved" });
assert.deepEqual(w.events("action.refused").map((e) => e.body.code), ["ref-moved"]);
assert.equal(w.gitea.requests.length, 0);
assert.equal(git(w.root, `--git-dir=${w.forkRepo}`, "for-each-ref", "refs/heads/coder/"), "");
});
test("T3: a push recorded while the bus process pushes refuses decision-mismatch; only the decision's commit reached the fork", async (t) => {
const w = await setup(t);
w.bus.credentials.issue("demo/coder", "gitea", TOKEN, "2099-01-01");
// The working repo's lock is held through the git push, so a socket push would wait for it; the
// broker's record of a newer push is what a push that won the lock leaves.
w.hooks.publishPush = async (m, next) => {
w.bus.checkPush(w.caps.coder, TASK);
w.bus.recordPush(w.caps.coder, { task_ref: TASK, project: "stack", branch: "feat", ref: workRef(TASK), commit: "c".repeat(40), previous: w.work });
return next();
};
assert.deepEqual(await w.publish(w.approve()), { ok: false, error: "decision-mismatch" });
assert.deepEqual(w.events("action.refused").map((e) => e.body.code), ["decision-mismatch"]);
assert.equal(w.events("action.allowed").filter((e) => e.body.action === "git.push.remote" && e.body.operation !== "decision.raise").length, 0);
assert.equal(git(w.root, `--git-dir=${w.forkRepo}`, "rev-parse", "refs/heads/coder/feat"), w.work);
assert.ok(w.logs.some((l) => l.includes("wasn't recorded")));
});
test("sage-1440: the publishResult poll is bounded: publish-timeout, filed once, and nothing recorded", async (t) => {
const w = await setup(t, { publishDeadlineMs: 150 });
let polls = 0;
w.hooks.publishResult = async () => {
polls++;
return { state: "running" };
};
assert.deepEqual(await w.publish(w.approve()), { ok: false, error: "publish-timeout" });
assert.ok(polls > 0);
assert.deepEqual(w.events("action.refused").map((e) => e.body.code), ["publish-timeout"]);
assert.equal(w.events("action.allowed").filter((e) => e.body.action === "git.push.remote" && e.body.operation !== "decision.raise").length, 0);
assert.ok(w.logs.some((l) => l.includes("publish-timeout")));
});
+3 -7
View File
@@ -101,10 +101,8 @@ export function gitEnv(env = process.env) {
// consequence: a Ctrl-C at the terminal of a host run by hand no longer
// reaches its git, though a systemd stop still does (it kills the cgroup).
// `stdin` names a file git reads as its input (row 137: patch-id); without
// it git gets none. `pass` (row 199, publish) is written once to a pipe on
// fd 3, which git and what it starts inherit; publish's credential helper
// reads its token there, so the token is in no argv, env, URL or file.
export function runGit(argv, env, limits = GIT_LIMITS, { stdin = null, pass = null } = {}) {
// it git gets none.
export function runGit(argv, env, limits = GIT_LIMITS, { stdin = null } = {}) {
return new Promise((resolve) => {
let input = "ignore";
try {
@@ -112,10 +110,8 @@ export function runGit(argv, env, limits = GIT_LIMITS, { stdin = null, pass = nu
} catch (e) {
return resolve({ ok: false, out: "", err: "", why: e.code ?? e.message });
}
const child = spawn("git", argv, { env, detached: true, stdio: [input, "pipe", "pipe", ...(pass === null ? [] : ["pipe"])] });
const child = spawn("git", argv, { env, detached: true, stdio: [input, "pipe", "pipe"] });
if (typeof input === "number") closeSync(input);
// A git that never asked for it closes the pipe unread: EPIPE is no fault.
if (pass !== null) child.stdio[3].on("error", () => {}).end(pass);
let out = "";
let err = "";
let signalled = false;
+1 -89
View File
@@ -63,22 +63,6 @@
// compare and swap (base-moved), then record() runs; when it throws, main
// goes back, and a main that can't go back refuses rollback-failed.
// dropLand removes the worktree; the caller runs it on every path.
//
// publish (row 199, #1697, design D9 to D9b) sends a recorded work commit to
// the agent's own fork. It runs in two processes, because the gitea token
// lives in the bus process and the lock in the host's:
// publish, in the host, under the lock: the seed; the work ref is at the
// source commit (ref-moved); every commit in main..source has author and
// committer email equal to the agent's (identity-mismatch); then
// remote(), which the caller wires to publishRemote in the bus process.
// publishRemote, in the bus process, with the token on fd 3 only:
// `ls-remote --symref <fork> HEAD <dst>`; a non-zero exit refuses
// fork-unavailable. The fork's HEAD must be a commit this repository has
// and an ancestor of main (remote-unrelated, an empty fork included).
// Then `push <fork> <source>:<dst>` by oid, with no `+`. Only https
// crosses (protocol.https.allow), redirects are refused, and output
// holding the token refuses credential-leak. Git's stderr is never
// returned: it can carry the URL.
import { randomBytes } from "node:crypto";
import { closeSync, lstatSync, mkdirSync, openSync, readSync, renameSync, rmSync, writeFileSync } from "node:fs";
@@ -117,7 +101,7 @@ export const FSCK = ["-c", "transfer.fsckObjects=true"];
// `limits` is { timeoutMs, graceMs }; the tests shorten them. runGit
// (checkout.mjs) runs git in its own process group and signals the group.
const git = (args, env, { pre = [], limits = GIT_LIMITS, stdin = null, pass = null } = {}) => runGit([...SAFE, ...pre, ...args], gitEnv(env), limits, { stdin, pass });
const git = (args, env, { pre = [], limits = GIT_LIMITS, stdin = null } = {}) => runGit([...SAFE, ...pre, ...args], gitEnv(env), limits, { stdin });
// git with this call's env and limits, for the helpers below.
const bound = (env, limits) => (args, opts = {}) => git(args, env, { ...opts, limits });
@@ -418,75 +402,3 @@ async function removeWorktree(repo, worktree, g) {
if (!r.ok) rmSync(worktree, { recursive: true, force: true });
await g([`--git-dir=${repo}`, "worktree", "prune"]);
}
// await publish({ dataRoot, business, project, root, taskRef, source, email, remote, env, limits })
// Resolves what remote() resolves; rejects with WorkingError.
export function publish({ dataRoot, business, project, root, taskRef, source, email, remote, env = process.env, limits = GIT_LIMITS }) {
if (!PROJECT.test(project ?? "") || !OID.test(source ?? "") || typeof email !== "string" || !email) refuse("invalid-request");
const ref = `refs/heads/${workRef(taskRef)}`;
const repo = workingRepo(dataRoot, business, project);
const g = bound(env, limits);
return locked(repo, async () => {
try {
if (!lstatSync(repo).isDirectory()) refuse("working-repo-invalid");
} catch (e) {
if (e instanceof WorkingError) throw e;
refuse("working-repo-invalid", `${repo} can't be read: ${e.code}`);
}
await seed(repo, root, g);
const at = `--git-dir=${repo}`;
if ((await tip(repo, ref, g)) !== source) refuse("ref-moved", `${ref} isn't at ${source}`);
const range = await g([at, "log", "--format=%ae%x00%ce", `refs/heads/main..${source}`]);
if (!range.ok) refuse("working-repo-invalid", range.why);
for (const line of range.out.split("\n").filter(Boolean)) {
if (line !== `${email}\0${email}`) refuse("identity-mismatch", `a commit in main..${source.slice(0, 12)} isn't by ${email}`);
}
return remote();
});
}
const HOST = /^[a-z0-9.-]{1,253}(:[0-9]{1,5})?$/;
const NAME = /^[A-Za-z0-9_.-]{1,100}$/;
// The one-shot helper: for a `get` on https at exactly the fork's host and
// port, it answers the user and the line it reads from fd 3. Anything else,
// and `store` and `erase`, gets no answer. HOST and NAME keep both shell-safe.
const helper = (host, user) =>
`!f() { test "$1" = get || exit 0; p=; h=; while IFS= read -r l; do case "$l" in protocol=*) p=\${l#protocol=};; host=*) h=\${l#host=};; "") break;; esac; done; ` +
`test "$p" = https && test "$h" = '${host}' || exit 0; IFS= read -r t <&3 || exit 0; printf 'username=%s\\npassword=%s\\n' '${user}' "$t"; }; f`;
// await publishRemote({ repo, fork, user, source, dst, token, tls, env, limits })
// Resolves { ref, commit, previous }, previous being dst's tip on the fork
// before the push, or null. `tls` is for the tests' scratch CA only.
export async function publishRemote({ repo, fork, user, source, dst, token, tls = [], env = process.env, limits = GIT_LIMITS }) {
let url;
try {
url = new URL(fork);
} catch {
refuse("invalid-request");
}
if (url.protocol !== "https:" || !HOST.test(url.host) || !NAME.test(user ?? "") || !OID.test(source ?? "") || !dst?.startsWith(`refs/heads/${user}/`)) refuse("invalid-request");
const secret = String(token).trim();
const net = async (args) => {
const pre = [...tls, "-c", "protocol.https.allow=always", "-c", "http.followRedirects=false", "-c", "credential.helper=", "-c", `credential.helper=${helper(url.host, user)}`];
const r = await git([`--git-dir=${repo}`, ...args], env, { pre, limits, pass: `${secret}\n` });
if (r.out.includes(secret) || r.err.includes(secret)) refuse("credential-leak");
return r;
};
const g = bound(env, limits);
const listed = await net(["ls-remote", "--symref", fork, "HEAD", dst]);
if (!listed.ok) refuse("fork-unavailable");
let head = null;
let previous = null;
for (const line of listed.out.split("\n")) {
const [oid, name] = line.split("\t");
if (!OID.test(oid ?? "")) continue;
if (name === "HEAD") head = oid;
else if (name === dst) previous = oid;
}
if (!head) refuse("remote-unrelated", "the fork is empty");
const known = await g([`--git-dir=${repo}`, "cat-file", "-e", `${head}^{commit}`]);
if (!known.ok || !(await g([`--git-dir=${repo}`, "merge-base", "--is-ancestor", head, "refs/heads/main"])).ok) refuse("remote-unrelated", "the fork's HEAD isn't in main's history");
const pushed = await net(["push", "--porcelain", "--no-follow-tags", fork, `${source}:${dst}`]);
if (!pushed.ok) refuse(/\((non-fast-forward|fetch first)\)/.test(pushed.out) ? "non-fast-forward" : "push-failed");
return { ref: dst, commit: source, previous };
}
-95
View File
@@ -1,95 +0,0 @@
// Row 199 (#1697, design T3): a fake Gitea for publish, on 127.0.0.1 over TLS with a scratch CA,
// serving bare repositories under `<dir>/repos/<user>/<repo>.git` through git http-backend. Every
// request needs Basic auth with the user and token it was made with, or with any pair in `users`,
// as a private repository on Gitea does. It records each request's method, path and user, never the password. `redirect`
// makes every request a 302 to `redirect`. A repository's own hooks run on a push, so a test can
// make the fork echo something back as `remote:` lines. Nothing here leaves 127.0.0.1.
import { spawn, spawnSync } from "node:child_process";
import { mkdirSync, readFileSync, writeFileSync } from "node:fs";
import { createServer } from "node:https";
import { join } from "node:path";
// A CA and a server certificate for IP 127.0.0.1, made once per directory with openssl.
export function scratchCa(dir) {
mkdirSync(dir, { recursive: true, mode: 0o700 });
const ssl = (...args) => {
const r = spawnSync("openssl", args, { cwd: dir, encoding: "utf8" });
if (r.status !== 0) throw new Error(`openssl ${args[0]}: ${r.stderr}`);
};
ssl("req", "-x509", "-newkey", "ec", "-pkeyopt", "ec_paramgen_curve:P-256", "-nodes", "-keyout", "ca.key", "-out", "ca.pem", "-days", "2", "-subj", "/CN=mosaic test CA");
ssl("req", "-newkey", "ec", "-pkeyopt", "ec_paramgen_curve:P-256", "-nodes", "-keyout", "server.key", "-out", "server.csr", "-subj", "/CN=127.0.0.1");
writeFileSync(join(dir, "san.ext"), "subjectAltName=IP:127.0.0.1\n");
ssl("x509", "-req", "-in", "server.csr", "-CA", "ca.pem", "-CAkey", "ca.key", "-CAcreateserial", "-out", "server.pem", "-days", "2", "-extfile", "san.ext");
return { ca: join(dir, "ca.pem"), key: readFileSync(join(dir, "server.key")), cert: readFileSync(join(dir, "server.pem")) };
}
export async function fakeGitea({ dir, tls, user, token, users = { [user]: token }, redirect = null }) {
const root = join(dir, "repos");
mkdirSync(root, { recursive: true });
const requests = [];
const server = createServer({ key: tls.key, cert: tls.cert }, (req, res) => {
const url = new URL(req.url, "https://127.0.0.1");
const auth = /^Basic (.+)$/.exec(req.headers.authorization ?? "");
const [who, pass] = auth ? Buffer.from(auth[1], "base64").toString().split(":") : [null, null];
requests.push({ method: req.method, path: url.pathname, user: who });
if (redirect) {
res.writeHead(302, { location: redirect + req.url });
return res.end();
}
if (!who || !Object.hasOwn(users, who) || pass !== users[who]) {
res.writeHead(401, { "www-authenticate": 'Basic realm="Gitea"' });
return res.end();
}
// /<owner>/<repo>/info/refs and the like, on <root>/<owner>/<repo>.git.
const m = /^\/([^/]+)\/([^/]+?)(?:\.git)?(\/.*)$/.exec(url.pathname);
if (!m) {
res.writeHead(404);
return res.end();
}
const cgi = spawn("git", ["http-backend"], {
env: {
PATH: process.env.PATH,
GIT_PROJECT_ROOT: root,
GIT_HTTP_EXPORT_ALL: "1",
GIT_CONFIG_NOSYSTEM: "1",
GIT_CONFIG_GLOBAL: "/dev/null",
HOME: "/nonexistent",
PATH_INFO: `/${m[1]}/${m[2]}.git${m[3]}`,
REQUEST_METHOD: req.method,
QUERY_STRING: url.search.slice(1),
CONTENT_TYPE: req.headers["content-type"] ?? "",
HTTP_CONTENT_ENCODING: req.headers["content-encoding"] ?? "",
GIT_PROTOCOL: req.headers["git-protocol"] ?? "",
REMOTE_USER: who,
REMOTE_ADDR: "127.0.0.1",
},
stdio: ["pipe", "pipe", "ignore"],
});
req.pipe(cgi.stdin);
const chunks = [];
cgi.stdout.on("data", (c) => chunks.push(c));
cgi.on("close", () => {
const out = Buffer.concat(chunks);
const split = out.indexOf("\r\n\r\n");
const head = out.subarray(0, split).toString().split("\r\n");
const headers = {};
let status = 200;
for (const line of head) {
const [k, ...v] = line.split(":");
if (k.toLowerCase() === "status") status = Number.parseInt(v.join(":").trim(), 10);
else headers[k] = v.join(":").trim();
}
res.writeHead(status, headers);
res.end(out.subarray(split + 4));
});
});
await new Promise((r) => server.listen(0, "127.0.0.1", r));
const port = server.address().port;
return {
port,
origin: `https://127.0.0.1:${port}`,
requests,
repo: (owner, name) => join(root, owner, `${name}.git`),
close: () => new Promise((r) => server.close(r)),
};
}
-185
View File
@@ -1,185 +0,0 @@
// Row 199 (#1697, design D9 step 3 to 5, D9a, T3): publish and publishRemote (src/working.mjs)
// against a TLS fake Gitea on 127.0.0.1 (fake-gitea.mjs) with a scratch CA, passed only through
// the `tls` parameter. Fake tokens in a scratch root. Each refusal that should reach the network
// checks the fake saw the request, so no case passes because git refused https first.
import { test } from "node:test";
import assert from "node:assert/strict";
import { spawnSync } from "node:child_process";
import { chmodSync, mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
import { tmpdir } from "node:os";
import { join } from "node:path";
import { gitEnv } from "../src/checkout.mjs";
import { publish, publishRemote, push, workingRepo } from "../src/working.mjs";
import { fakeGitea, scratchCa } from "./fake-gitea.mjs";
const TASK = "vikunja:1/2";
const TOKEN = "fake_publish_token_0123456789";
const USER = "coder";
const EMAIL = "[email protected]";
const DST = "refs/heads/coder/feat";
function git(cwd, ...args) {
const r = spawnSync("git", args, { cwd, encoding: "utf8", env: gitEnv() });
assert.equal(r.status, 0, `git ${args.join(" ")}: ${r.stderr}`);
return r.stdout.trim();
}
function commit(dir, text, email = EMAIL) {
writeFileSync(join(dir, "README"), `${text}\n`);
git(dir, "add", "README");
git(dir, "-c", `user.name=${text}`, "-c", `user.email=${email}`, "commit", "-q", "-m", text);
return git(dir, "rev-parse", "HEAD");
}
const rejects = (p, code) => assert.rejects(p, (e) => e.code === code || assert.fail(`${e.code}: ${e.message}, wanted ${code}`));
// A root, a coder's clone with work on feat pushed to the working repo, and a fake Gitea whose
// coder/stack fork is a bare clone of the root.
async function world(t, { fork = true, redirect = null } = {}) {
const dir = mkdtempSync(join(tmpdir(), "mosaic-publish-"));
t.after(() => rmSync(dir, { recursive: true, force: true }));
const root = join(dir, "root");
mkdirSync(root);
git(root, "init", "-q", "-b", "main");
const first = commit(root, "one", "[email protected]");
const clone = join(dir, "clone");
git(dir, "clone", "-q", root, clone);
git(clone, "checkout", "-q", "-b", "feat");
const work = commit(clone, "work");
const dataRoot = join(dir, "data");
mkdirSync(dataRoot, { mode: 0o700 });
const file = join(dir, "b.bundle");
git(clone, "bundle", "create", "-q", file, "refs/heads/feat", "--not", "--remotes=origin");
await push({ dataRoot, business: "acme", project: "stack", root, taskRef: TASK, branch: "feat", bundle: file });
const ca = scratchCa(join(dir, "ca"));
const gitea = await fakeGitea({ dir: join(dir, "gitea"), tls: ca, user: USER, token: TOKEN, redirect });
t.after(() => gitea.close());
const forkRepo = gitea.repo(USER, "stack");
if (fork) {
mkdirSync(join(forkRepo, ".."), { recursive: true });
git(dir, "clone", "-q", "--bare", root, forkRepo);
}
const tls = ["-c", `http.sslCAInfo=${ca.ca}`];
const repo = workingRepo(dataRoot, "acme", "stack");
const remote = (extra = {}) => publishRemote({ repo, fork: `${gitea.origin}/${USER}/stack`, user: USER, source: work, dst: DST, token: TOKEN + "\n", tls, ...extra });
const go = (extra = {}) => publish({ dataRoot, business: "acme", project: "stack", root, taskRef: TASK, source: work, email: EMAIL, remote: () => remote(), ...extra });
return { dir, root, first, clone, work, dataRoot, repo, gitea, forkRepo, tls, remote, go };
}
test("T3: publish pushes the work commit by oid to <user>/<branch> on the fork, with the token only through the helper", async (t) => {
const w = await world(t);
assert.deepEqual(await w.go(), { ref: DST, commit: w.work, previous: null });
assert.equal(git(w.dir, `--git-dir=${w.forkRepo}`, "rev-parse", DST), w.work);
// Every request went to the fork's path, and each one that carried auth carried the agent's user.
assert.ok(w.gitea.requests.length > 0);
for (const r of w.gitea.requests) assert.match(r.path, /^\/coder\/stack\/(info\/refs|git-receive-pack|git-upload-pack)$/);
assert.ok(w.gitea.requests.some((r) => r.user === USER && r.path.endsWith("git-receive-pack")));
// A second publish of the same commit is a no-op push and names the previous tip.
assert.deepEqual(await w.go(), { ref: DST, commit: w.work, previous: w.work });
// The working repo holds no remote and no credential config.
assert.equal(spawnSync("git", [`--git-dir=${w.repo}`, "config", "--get-regexp", "^(remote|credential)\\."], { env: gitEnv() }).stdout.toString(), "");
});
test("T3: a moved fork branch refuses non-fast-forward; no + and no force", async (t) => {
const w = await world(t);
const other = join(w.dir, "other");
git(w.dir, "clone", "-q", w.root, other);
const elsewhere = commit(other, "elsewhere");
git(other, "push", "-q", w.forkRepo, `${elsewhere}:${DST}`);
await rejects(w.go(), "non-fast-forward");
assert.equal(git(w.dir, `--git-dir=${w.forkRepo}`, "rev-parse", DST), elsewhere);
});
test("T3: D9a: an empty fork and an unrelated fork refuse remote-unrelated; a missing fork refuses fork-unavailable", async (t) => {
const empty = await world(t, { fork: false });
git(empty.dir, "init", "-q", "--bare", empty.forkRepo);
await rejects(empty.go(), "remote-unrelated");
assert.ok(empty.gitea.requests.length > 0);
assert.equal(empty.gitea.requests.some((r) => r.path.endsWith("git-receive-pack")), false);
const unrelated = await world(t, { fork: false });
const alone = join(unrelated.dir, "alone");
mkdirSync(alone);
git(alone, "init", "-q", "-b", "main");
commit(alone, "unrelated");
git(unrelated.dir, "clone", "-q", "--bare", alone, unrelated.forkRepo);
await rejects(unrelated.go(), "remote-unrelated");
assert.equal(unrelated.gitea.requests.some((r) => r.path.endsWith("git-receive-pack")), false);
const missing = await world(t, { fork: false });
await rejects(missing.go(), "fork-unavailable");
assert.ok(missing.gitea.requests.length > 0);
});
test("T3: a fork past main refuses remote-unrelated until root catches up", async (t) => {
const w = await world(t);
const other = join(w.dir, "other");
git(w.dir, "clone", "-q", w.forkRepo, other);
const ahead = commit(other, "ahead");
git(other, "push", "-q", w.forkRepo, `${ahead}:refs/heads/main`);
await rejects(w.go(), "remote-unrelated");
});
test("T3: a wrong token refuses fork-unavailable; git's stderr isn't in the refusal", async (t) => {
const w = await world(t);
await assert.rejects(w.remote({ token: "fake_wrong_token_0123456789" }), (e) => e.code === "fork-unavailable" && e.message === "fork-unavailable");
assert.ok(w.gitea.requests.length > 0);
assert.equal(git(w.dir, `--git-dir=${w.forkRepo}`, "for-each-ref", DST), "");
});
test("T3: a 302 is not followed: fork-unavailable, and the target sees no request", async (t) => {
const w = await world(t);
const target = await fakeGitea({ dir: join(w.dir, "target"), tls: scratchCa(join(w.dir, "ca2")), user: USER, token: TOKEN });
t.after(() => target.close());
const bounce = await fakeGitea({ dir: join(w.dir, "bounce"), tls: scratchCa(join(w.dir, "ca3")), user: USER, token: TOKEN, redirect: target.origin });
t.after(() => bounce.close());
const tls = ["-c", `http.sslCAInfo=${join(w.dir, "ca3", "ca.pem")}`];
await rejects(w.remote({ fork: `${bounce.origin}/${USER}/stack`, tls }), "fork-unavailable");
assert.ok(bounce.requests.length > 0);
assert.equal(target.requests.length, 0);
});
test("T3: the helper answers only the fork's host and port: a fork on another port gets no token", async (t) => {
const w = await world(t);
// The fork names port 1; insteadOf (through the tests' tls parameter) sends git to the fake's
// port. The helper was told port 1, so it answers nothing and no password goes out.
const before = w.gitea.requests.length;
await rejects(w.remote({ fork: `https://127.0.0.1:1/${USER}/stack`, tls: [...w.tls, "-c", `url.https://127.0.0.1:${w.gitea.port}/.insteadOf=https://127.0.0.1:1/`] }), "fork-unavailable");
assert.ok(w.gitea.requests.length > before, "the rewritten URL reached the fake");
assert.equal(w.gitea.requests.slice(before).some((r) => r.user !== null), false, "no credential went to a host the helper wasn't told");
});
test("T3: a fork that echoes the token refuses credential-leak", async (t) => {
const w = await world(t);
const hook = join(w.forkRepo, "hooks", "pre-receive");
writeFileSync(hook, `#!/bin/sh\necho ${TOKEN} >&2\nexit 0\n`);
chmodSync(hook, 0o755);
await rejects(w.go(), "credential-leak");
});
test("T3: no tls parameter, no trust: the scratch CA reaches git only through it", async (t) => {
const w = await world(t);
await rejects(w.remote({ tls: [] }), "fork-unavailable");
});
test("D9 steps 3 and 4: ref-moved and identity-mismatch refuse before any request", async (t) => {
const w = await world(t);
await rejects(w.go({ source: w.first }), "ref-moved");
const other = commit(w.clone, "by someone else", "[email protected]");
const file = join(w.dir, "b2.bundle");
git(w.clone, "bundle", "create", "-q", file, "refs/heads/feat", "--not", "--remotes=origin");
await push({ dataRoot: w.dataRoot, business: "acme", project: "stack", root: w.root, taskRef: TASK, branch: "feat", bundle: file });
await rejects(w.go({ source: other }), "identity-mismatch");
assert.equal(w.gitea.requests.length, 0);
});
test("publishRemote refuses a fork, user or destination that isn't shell- and ref-safe before any git", async (t) => {
const w = await world(t);
for (const extra of [
{ fork: `http://127.0.0.1:${w.gitea.port}/${USER}/stack` },
{ fork: `https://127.0.0.1:${w.gitea.port}'/${USER}/stack` },
{ user: "co'der" },
{ dst: "refs/heads/main" },
{ dst: "refs/heads/pm/feat" },
{ source: "HEAD" },
]) await rejects(w.remote(extra), "invalid-request");
assert.equal(w.gitea.requests.length, 0);
});
+2 -6
View File
@@ -80,11 +80,7 @@ export function tasksAdapter({
tree: null,
bots: new Set([businesses[name].tracker.sync.botId, ...Object.values(s.roles).map((r) => r.botId)]),
sync: use('@sync'),
// Row 199 (#1697, design D5): a request's call keys from the broker's session for its cap,
// never from a role name. probe is the host's own start check of each role's token.
as: (session, method, path, opts) =>
credentials.useFor(session, 'vikunja', (token) => call(token, method, path, opts)),
probe: (role, method, path, opts) => use(role)(method, path, opts),
as: (role, method, path, opts) => use(role)(method, path, opts),
state: 'starting',
refused: null,
queue: Promise.resolve(),
@@ -100,7 +96,7 @@ export function tasksAdapter({
timers: [],
};
ctx.syncCall = (_token, method, path, opts) => ctx.sync(method, path, opts);
ctx.asCall = (role) => (_token, method, path, opts) => ctx.probe(role, method, path, opts);
ctx.asCall = (role) => (_token, method, path, opts) => ctx.as(role, method, path, opts);
ctx.handle = verbs(ctx);
contexts.set(name, ctx);
}
+2 -2
View File
@@ -41,7 +41,7 @@ export async function startup(ctx) {
if (probe.status === 404 || probe.status === 403 || outcome(probe) === 'ok') refuse('scope-too-broad');
expect(probe, 401, 'tracker-unexpected');
for (const [role, r] of Object.entries(ctx.roles)) {
const control = await ctx.probe(role, 'GET', `/tasks/${MISSING}`);
const control = await ctx.as(role, 'GET', `/tasks/${MISSING}`);
if (control.status !== 404 || control.json?.code !== 4002) {
expect(control, 404, 'tracker-unexpected');
refuse('tracker-unexpected');
@@ -49,7 +49,7 @@ export async function startup(ctx) {
const probes = [['DELETE', `/tasks/${MISSING}`, undefined]];
if (r.definition !== 'pm') probes.push(['POST', `/tasks/${MISSING}/labels`, { label_id: 1 }]);
for (const [method, path, body] of probes) {
const p = await ctx.probe(role, method, path, { body });
const p = await ctx.as(role, method, path, { body });
if (p.status === 404 || p.status === 403 || outcome(p) === 'ok') refuse('scope-too-broad');
if (p.status !== 401) refuse(p.status === 0 || p.status >= 500 ? 'tracker-unavailable' : 'tracker-unexpected');
}
+8 -9
View File
@@ -122,9 +122,8 @@ export function verbs(ctx) {
const relatedTo = (task, kind, other) =>
Array.isArray(task.related_tasks?.[kind]) && task.related_tasks[kind].some((t) => t?.id === other);
// One write with the acting role's token. An uncertain answer is settled by a re-read.
// `me` is the broker's session for the caller's cap (design D5).
async function step(me, method, path, body, at, done) {
const r = await ctx.as(me, method, path, { body });
async function step(role, method, path, body, at, done) {
const r = await ctx.as(role, method, path, { body });
const o = outcome(r);
if (o === 'ok') return r;
if (o === 'uncertain' && at !== null && done) {
@@ -145,7 +144,7 @@ export function verbs(ctx) {
if (!mine) fail('tracker-role');
if (PM_VERBS.has(verb) && mine.definition !== 'pm') fail('field-writer');
const plan = prepare(verb, args, role, mine);
if (verb === 'task.create') return create(cap, me, plan);
if (verb === 'task.create') return create(cap, role, plan);
const at = target(args.task_ref);
const before = await read(at);
if (args.expect !== undefined && args.expect !== before.digest) {
@@ -166,7 +165,7 @@ export function verbs(ctx) {
let landed = 0;
try {
for (const [method, path, body, done] of work.steps) {
await step(me, method, path, body, at, done);
await step(role, method, path, body, at, done);
landed++;
}
} catch (e) {
@@ -386,12 +385,12 @@ export function verbs(ctx) {
fail('unknown-verb');
}
}
async function create(cap, me, plan) {
async function create(cap, role, plan) {
if (!ctx.broker.taskView(ctx.business, 'input', plan.request)) fail('request-not-found');
await located(plan.others);
ctx.broker.authorize(cap, 'task.create', { decision: plan.decision, target: null });
// No re-read can find a create whose answer was lost, so an uncertain create is reported as such.
const r = await step(me, 'POST', `/projects/${plan.at.project}/tasks`, plan.body, null, null);
const r = await step(role, 'POST', `/projects/${plan.at.project}/tasks`, plan.body, null, null);
if (r.status !== 201 && r.status !== 200) fail('tracker-unexpected');
checkTask(r.json);
const id = r.json.id;
@@ -400,9 +399,9 @@ export function verbs(ctx) {
let failure = null;
try {
for (const l of plan.labels)
await step(me, 'POST', `/tasks/${id}/labels`, { label_id: l }, at, (a) => a.fields.labels.includes(l));
await step(role, 'POST', `/tasks/${id}/labels`, { label_id: l }, at, (a) => a.fields.labels.includes(l));
for (const x of plan.relations)
await step(me, 'POST', `/tasks/${id}/relations`, { other_task_id: x.other, relation_kind: x.kind }, at, (a) =>
await step(role, 'POST', `/tasks/${id}/relations`, { other_task_id: x.other, relation_kind: x.kind }, at, (a) =>
relatedTo(a.task, x.kind, x.other),
);
} catch (e) {
+1 -2
View File
@@ -73,8 +73,7 @@ business's arbiter for its domain. Pass `domain` to `raise_decision`,
gated raise, and every hard wait, goes to the human through the inbox.
The hard waits are `spend`, `data.delete`, `network.expose`,
`account.create`, `credential.mint`, `message.external`,
`git.push.protected`, `git.push.remote` and `deploy`; a citation never
moves them.
`git.push.protected` and `deploy`; a citation never moves them.
A `role.revoke` raise whose target is the assistant is a hard wait too:
it goes to the human, cited or not.
+1 -2
View File
@@ -53,8 +53,7 @@ business's arbiter for its domain. Pass `domain` to `raise_decision`,
gated raise, and every hard wait, goes to the human through the inbox.
The hard waits are `spend`, `data.delete`, `network.expose`,
`account.create`, `credential.mint`, `message.external`,
`git.push.protected`, `git.push.remote` and `deploy`; a citation never
moves them.
`git.push.protected` and `deploy`; a citation never moves them.
A `role.revoke` raise whose target is the assistant is a hard wait too:
it goes to the human, cited or not.
+1 -1
View File
@@ -35,6 +35,6 @@ if [ "${1:-}" = business ]; then
exec node "$REPO/packages/business/src/cli.mjs" "$@"
fi
case "${1:-}" in
usage|project|limits|inbox|decide|tasks|agents|trail|talk|stop|launches|bus|plugins|credentials|name|names|follow|threads|thread) exec node "$REPO/packages/cli/src/cli.mjs" "$@" ;;
usage|project|limits|inbox|decide|tasks|agents|trail|talk|stop|launches|bus|plugins|name|names|follow|threads|thread) exec node "$REPO/packages/cli/src/cli.mjs" "$@" ;;
esac
exec node "$REPO/packages/seat/src/cli.mjs" "$@"