stack/agents/darkwing/work/queue-195/DESIGN.md

# Row 195 design: the Case 1 install on 10.1.1.53 (r7)

Darkwing, 2026-10-21. Brief: `docs/plans/2026-10-20_hub-own-uid.md`, blob 45be7f09afc1, section "Brief: the
Case 1 install on 10.1.1.53 (row 195)". Gitea #1693. Reviewer Duckworth, gate owner Sage. Branch
`darkwing/row195-r7` on c515273db. r6 (b7239bb66 on `darkwing/row195`, approved duckworth-0055) is the base,
carried unchanged as 7398b7dd7. The r6 text is in that commit and in b7239bb66.

## Changes from r6

Q187 (mos-1108) made install and upgrade product commands, and mos-1118 split r6 into the installer rows
(`docs/plans/2026-10-20_installer-rows.md`). That brief says row 195 "shrinks to the factory profile's
values and the Case 1 acceptance". So r7 is mostly a cut. What r6 designed now ships elsewhere:

| r6 piece | Now | State |
|---|---|---|
| `stack-release.sh --units external`: `confirm`, `revert`, `gate`, `rollback --to`, external `adopt` and `status`; `stack-copy.sh check --units external` and bundle copies | row 232, I1 | landed |
| `case1-repin.sh` (A1-A5, A2f, R1-R4) | `scripts/stack-upgrade.sh`, row 233, I3 | landed |
| the host steps 2a, 2a', 2a'' and the window's root steps | `docs/guides/host-run-sheet.md` sections 1-10, row 233 | landed |
| `case1-hostcheck.sh`, `case1-pins.sh`, `mosaic-hostcheck.service` (the drift stop, r6 B2) | row 240, I4 | landed c515273db |
| the unit render, `--profile factory`, `--pm`, `--claude`, the factory serve flags, the precheck | row 230, C2-13a, I2 | landed |
| the install as `mosaic`: bootstrap, business file, copy | row 226, C2-13b, I2, `scripts/case2-install.sh` | assigned to Gosalyn, briefed, not started |
| `docs/guides/case1-install.md` | the host run sheet, plus section 6 of this file | |

Other changes:

- Claude moves under `/opt` (sage-1522, run sheet section 3). `--claude` is
  `/opt/mosaic-stack/claude-<pin>/node_modules/@anthropic-ai/claude-code/bin/claude.exe`, root's, and root
  installs and upgrades it. r6's `/srv/mosaic/.local/bin/claude` is gone. F6 reads the new path.
- The args file is `/etc/mosaic-stack/dev.args` (run sheet section 7), not `case1-unit.args`. It holds
  `--copy` too, and `stack-upgrade.sh` rewrites only that line, after confirm passes.
- Both units are `WantedBy=multi-user.target` and start at boot (Q189, mos-1122). r6's "the Console has no
  `[Install]`, root starts it by hand after a reboot" no longer holds.
- The business id is stated, with the DEFERRED 240 N1 reasoning (sage-1809, sage-1813). Section 7.
- The window can't run until row 226 lands. Sage added 226 to row 195's after list at rev 2718
  (sage-1813). Section 6 lists what row 195 needs from 226's install.
- The acceptance runs as root through ops-01, not as `localadmin`. `aa-status`, `sudo -l -U`,
  `systemd-run --uid` and the hostcheck all need root. It gains four functional checks for items the run
  sheet marks "precheck line missing" (F8-F11) and one host line (P11). Section 5.
- The Discord row in section 2 changed. The precheck's "no Discord binding" line is C2-13b's and is
  library-only, so for the factory it's F11 in the acceptance.

What row 195 still builds: `scripts/case1-accept.sh` and its test (section 5). The rest of this file is
inputs, the window order and the record the reviewer checks.

## 0. What I checked on the host

Read only, as `localadmin` over ssh, 2026-10-20 between 20:40Z and 21:51Z. Nothing in r7 touched .53
(sage-1809). The window's first step re-reads these, since they can drift in a day: unattended-upgrades
runs daily.

| Item | 10.1.1.53 on 2026-10-20 | 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 | run sheet section 1 |
| node | `/usr/local/bin/node` v22.22.2, root-owned | 24 or later, under `/opt` (run sheet section 2) |
| claude | not on localadmin's PATH | under `/opt` (run sheet section 3) |
| bwrap | absent; apt candidate `bubblewrap 0.9.0-1ubuntu0.3` | run sheet section 4 |
| user namespaces | `kernel.apparmor_restrict_unprivileged_userns = 1`, apparmor 4.0.1, no bwrap profile loaded | run sheet section 4 |
| profiles that name bwrap | none (`grep -rlw bwrap /etc/apparmor.d` exits 1) | hostcheck item 3 |
| `fs.protected_hardlinks` | 1 | hostcheck item 6 |
| `net.ipv4.ip_unprivileged_port_start` | 1024 | run sheet section 5 |
| listeners below 1024 | 22 (sshd) and 53 (the resolver); 732 is free | the Console's port |
| ports 7330-7333 | free | 7333 is the tester port |
| `localadmin` | NOPASSWD sudo (#423 step 1, mos-1066, mos-1077) | ops-01's root path |
| `jarvis` | groups sudo, docker, lxd, adm | #423 step 4 removes them (P5) |
| unattended-upgrades | on, with the security pocket, which carries bubblewrap | the drift stop (row 240) |
| journald | `SplitMode=uid` | the PM check reads `launches list`, not the journal (row 232) |

## 1. The factory profile's values

Instance `dev`, row 190's default (mos-1118 choice 4). `~` is `mosaic`'s home.

| Input | Value | Who gives it |
|---|---|---|
| user | `mosaic`, primary group `mosaic` with no other member | fixed |
| home | `/srv/mosaic`, 0700 | fixed |
| config dir | `/srv/mosaic/.config/mosaic-dev` | row 190 |
| data root | `/srv/mosaic/.mosaic-dev`, marker `{"instanceVersion":1,"name":"dev"}` | row 190 |
| business | `mosaic-stack`, file `<configDir>/businesses/mosaic-stack.json` | fixed (section 7) |
| profile | `factory` for `case2-install.sh`, `case2-unit.sh` and `case2-precheck.sh` | fixed |
| roles | the factory set from `roles/`, with a chief that has a harness var | the go names the set (section 2) |
| human id | Jason's, not `tester` (`broker.mjs:321`) | the go |
| model credential | `<configDir>/secrets/claude-session.env`, `mosaic` 0600, one line `CLAUDE_CODE_OAUTH_TOKEN=...` | Jason places it (section 3) |
| token slot path | none | section 2 |
| node | `/opt/mosaic-stack/node-<v>/bin/node`, `<v>` 24 or later | the go names `<v>` and its sha256 |
| claude | `/opt/mosaic-stack/claude-<pin>/node_modules/@anthropic-ai/claude-code/bin/claude.exe`, `<pin>` 2.1.304 unless the go names a newer one | the go |
| PM | on (`--pm`) | fixed |
| Console | `--port 732` on 127.0.0.1, with the unit's `CAP_NET_BIND_SERVICE` | fixed |
| tester | `--tester-port 7333` on 127.0.0.1 (P10) | fixed |
| board | none | fixed |
| release | a git bundle of `refs/heads/refactor` and the 40-hex commit (run sheet section 6) | the go names both and the bundle's sha256 |

The args file. `/etc/mosaic-stack/dev.args`, `root:root` 0644, one argument per line (run sheet section 7):

```
--profile
factory
--user
mosaic
--home
/srv/mosaic
--business
mosaic-stack
--copy
<40-hex commit>
--node
/opt/mosaic-stack/node-<v>/bin/node
--claude
/opt/mosaic-stack/claude-<pin>/node_modules/@anthropic-ai/claude-code/bin/claude.exe
--pm
--
--port
732
--business
mosaic-stack
--tester-port
7333
```

`case2-unit.sh` takes exactly these before `--` (`scripts/case2-unit.sh:54`), and `stack-upgrade.sh`
refuses `args-unreadable` on anything else (`scripts/stack-upgrade.sh:126-148`). The factory profile
refuses the preview trio, `--send` and the proof pair (`case2-unit.sh:169`). .53 has no proxy, and Jason
reaches the Console with `ssh -L` (sage-1451 Q3). Changing a serve flag is a re-install with its own go,
not an upgrade.

Why the port is 732 (r6, mos-1101 item 5). Role sessions share the host's network namespace
(`packages/seat/src/sandbox.mjs:67`), so while the Console is down a session could bind a port above the
floor and answer in its place. Below 1024 only the Console unit can bind, through its capability. The
tester port can't go below 1024 (`packages/webui/src/cli.mjs:33`), so P10 records it.

PATH for `mosaic`'s steps. `sudo -u mosaic -H` resets PATH to sudo's `secure_path`, which finds
`/usr/local/bin/node`, v22. So every `mosaic` step runs as
`sudo -u mosaic -H env PATH=/opt/mosaic-stack/node-<v>/bin:/usr/local/bin:/usr/bin:/bin ...`, the list
`case2-unit.sh` renders as the units' PATH. `stack-upgrade.sh` builds it from the args file's `--node`.

## 2. Where Case 1 differs from Case 2

| Case 2 (library) | Case 1 (factory) | Reason |
|---|---|---|
| business file with the library mapping | `mosaic-stack` with no `library` key. The precheck refuses a library project under `factory` (`profile-library-business`, `case2-precheck.sh:127-128`) | the factory has no books repo |
| `bean`, `chet`, persona vars and files, specialists | the factory roles from `roles/`, with a chief that has a harness var | the go names the set. I haven't read Jason's live business file and won't. The acceptance needs only the chief (F3); a chief with no harness var answers `no-route`. The human adds roles later 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) |
| `bus.tester` off | on; the precheck refuses it off under `factory` (`case2-precheck.sh:135`) | the acceptance and the Q173 test system need the tester |
| persona modes and digests, `library.sync`, `library.status` | skipped | no persona files, no library (C2-13b brief, `factory` profile) |
| `ZAI_API_KEY` in `claude-session.env` | not checked; the file holds `CLAUDE_CODE_OAUTH_TOKEN` | no instance has provider `zai` (run sheet section 9) |
| no Discord binding (C2-13b, library only) | none either. F11 checks that `<dataRoot>/discord/` holds no binding | the factory's connector stays on dragon-lin. Two connectors on one bot token would split its events |
| the claude marker prompt | runs, since the chief's harness is claude-code | |
| serve flags: preview trio, `--send`, proof pair | `--port 732`, `--business mosaic-stack`, `--tester-port 7333` | no proxy on .53 |
| bus: no `--pm`, no `MOSAIC_CLAUDE_BIN` | `--pm`, and `Environment=MOSAIC_CLAUDE_BIN=<--claude>` | the PM dispatches the factory's work; without the path the bus finds no claude on the unit's PATH |
| Console proof file (run sheet section 9, .54) | none | no proof pair |

All of these are inputs to the landed scripts. Row 230 implemented the four factory items r6 asked for
(serve flags, `--pm`, `--claude`, row 190's tokens), and nothing in them hardcodes a Case 1 value.

## 3. Model access

Q185: Case 1 uses the account the factory uses today, the Claude subscription behind dragon-lin's
`claude-session.env`. No script copies that file and none reads its value. 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 (run sheet
section 9). Nobody prints it. The check is `stat -c '%U %a %F' PATH` only (F8).

The bus unit reads it through `EnvironmentFile=-` (`case2-unit.sh:243`). The leading `-` means a missing
file doesn't stop the bus, so the chief's first launch would fail instead. The window places the file
before the units start (section 6, W9), and the precheck's claude marker prompt then proves it works
without printing it.

Quota: the factory on dragon-lin and on .53 share one subscription. Q185's "use existing" accepts that.
The pi harness stays off on .53 until P19 #383 settles (mos-0865). The business file names claude only.

## 4. Upgrade, rollback and backup on .53

All three are `scripts/stack-upgrade.sh` (row 233), run as root by ops-01 from `/root/mosaic-stack` at the
reviewed commit. Usage, exits and the step log are in `docs/TOOLS.md`, "Root's upgrade driver", and run
sheet section 6.

| Action | Command | Who | Go |
|---|---|---|---|
| upgrade | `stack-upgrade.sh upgrade --instance dev --bundle /srv/mosaic/incoming/<sha>.bundle --sha <sha>` after run sheet section 6's per-upgrade lines | ops-01 | Mos per run; `--migrate` also needs Jason present (decision 96) |
| rollback | `stack-upgrade.sh rollback --instance dev --to <copy> [--restore <ACT> [--events N]]` | ops-01 | Mos per run. Mos or Sage picks `<copy>` from `stack-release.sh status mosaic-stack --units external`, read as `mosaic`. The plan refuses unless it is the previous distinct copy |
| backup of the store | A3's `backup_store` into `<dataRoot>/backups/stack/<ACT>`, inside every upgrade | the driver | the upgrade's |
| rehearsal | `stack-release.sh rehearse` against the log's newest backup, before anything stops, on `--rehearse-port`, default 7339, which section 0 found free | the driver | the upgrade's |
| backup of the VM | a PBS backup of VM 1008 before the install window and before any `--migrate` | ops-01 | the window's go (Q182, Q183: agents keep PBS) |

Three .53 specifics:

1. The first upgrade after install needs `--no-rehearsal` in its go. The install's adopt takes no backup,
   so the log records none, the rehearsal exits 3 (`no-backup`), and the driver goes on only with the
   flag (`stack-upgrade.sh:323-324`). Every later upgrade refuses the flag (`rehearsal-required`).
2. Exit 5 (`args-failed`): the new stack is live and confirmed, but `dev.args` still names the old copy.
   ops-01 moves the file the result line names into place with `mv -f`, and nothing else (run sheet
   section 6).
3. Exit 2 leaves both units stopped and the activation open. ops-01 sends Mos the exit and the last
   lines of `/var/log/mosaic-stack/upgrade-dev.log`, and Jason or Mos settles it with `confirm` or `revert`
   as on dragon-lin. A unit that doesn't stop within `--timeout` ends here on purpose (A2f's
   `stale-start`, duckworth-0055 N1, now in run sheet section 6).

The store backup doesn't hold the config dir or the credential. PBS backups of VM 1008 hold the whole disk,
the credential included (P4). A restore swaps only the store and leaves `claude-session.env` in place.

Re-install, for a serve flag or role change, is not an upgrade. It needs its own go and runs: stop the
Console, then the bus; rewrite `dev.args` with the same `--copy`; `case2-unit.sh` with it; both units start
again from the render. Row 230 N1 (webby-0046) holds here: a re-render with the units running leaves each on
its old copy, so ops-01 stops both first. `stack-upgrade.sh` already stops both before it renders.

## 5. Acceptance: `scripts/case1-accept.sh`

New in this row. ops-01 runs it as root on .53, from `/root/mosaic-stack`:

```
case1-accept.sh --business mosaic-stack --args /etc/mosaic-stack/dev.args --claude-pin <pin> --node-pin <v>
```

It reads the user, home, node, claude and ports from the args file and refuses (exit 2) when that file
fails `stack-upgrade.sh`'s rules. Every check inside the instance runs as
`sudo -u mosaic -H env PATH=<node dir>:/usr/local/bin:/usr/bin:/bin`. It changes nothing on the host
except what F3's message writes to the store. It prints one line per check, `<id> <path> <mark>
<evidence>`, then a count of each mark. Marks: `closed`, `accepted Q182`, `accepted Q183`, `accepted
mos-1096`, `accepted mos-1101`, and `unnamed`, which means "not named in Q182, Q183 or a Mos ruling".
It exits 1 when any F check fails and 0 otherwise. A P check never changes the exit. Each `unnamed`
line goes to Mos, who takes it to Jason (R8 N1). Duckworth reads the whole output.

It prints no value from `claude-session.env` and never opens it.

Functional checks. Any failure fails the row:

| Id | Check | How |
|---|---|---|
| F1 | the units run as `mosaic` | for `[email protected]` and `mosaic-webui.service`: active, `User=mosaic`, and the uid in `/proc/<MainPID>/status` is `mosaic`'s |
| F2 | the Console answers on 127.0.0.1:732 only | `curl -fsS http://127.0.0.1:732/healthz` succeeds, and `ss -Hltnp` shows 732 bound to 127.0.0.1 only, by the Console's MainPID. Jason reaches it with `ssh -L 7332:127.0.0.1:732 [email protected]`. A LAN listener waits for Q175 SSO (sage-1451 Q3) |
| F3 | one tester message gets a reply | as `mosaic`: `mosaic talk chief <marker> --business mosaic-stack --tester --wait 300`. A stored reply to the marked message passes. The inbox shows decisions only, so it isn't used |
| F4 | the precheck passes | `case2-precheck.sh --business mosaic-stack --profile factory` as `mosaic` exits 0 |
| 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 child. Its `/proc/<pid>/attr/current` equals `bwrap (enforce)` and the child's equals `bwrap//&unpriv_bwrap (enforce)`, exact strings after the trailing newline is cut |
| F6 | claude is the pin | the args file's `--claude` path sits under `/opt/mosaic-stack/claude-<pin>/`, and its `--version` as `mosaic` prints `<pin> (Claude Code)` |
| F7 | the PM run is live | `launches list --json` as `mosaic` has exactly one live entry for the PM's instance (`launch.by`), the check `confirm` runs |
| F8 | the credential's file | `stat -c '%U %a %F'` on `claude-session.env` reads `mosaic 600 regular file`, and it isn't a symlink. Run sheet section 9 marks this "precheck line missing" |
| F9 | the autoupdater is off | `/etc/claude-code/managed-settings.json` is `root:root` 0644 and its `env.DISABLE_AUTOUPDATER` is `"1"`. Run sheet section 3 marks this "precheck line missing" |
| F10 | node is the pin | the args file's `--node` is `/opt/mosaic-stack/node-<v>/bin/node` with `<v>` the `--node-pin`, and its `--version` prints `<v>`. Run sheet section 2 marks this "precheck line missing" |
| F11 | no Discord binding | `/srv/mosaic/.mosaic-dev/discord/` is absent or holds no `*.json` (section 2) |

Host paths. These report and never fail the row:

| Id | Path | Mark | Evidence |
|---|---|---|---|
| P1 | `mosaic` has sudo | closed only when `sudo -l -U mosaic` prints "is not allowed to run sudo"; 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 | `sudo -l -U localadmin` |
| P4 | root on pve over VM 1008, and PBS backups that hold `/srv/mosaic` with the credential | accepted Q182 and Q183 | a fixed line; the script can't see pve |
| P5 | `jarvis`, which runs Hermes, reaches `/srv/mosaic` through sudo, docker or lxd | closed when `jarvis` is in none of the three; unnamed otherwise. #423 step 4 removes all four groups before the install (mos-1091 4c) | `id jarvis` |
| P6 | other members of sudo, docker, lxd | closed when sudo holds only `localadmin` and docker and lxd are empty; each other member is its own unnamed line | `getent group` lines |
| P7 | a file under `/etc/sudoers.d` naming `mosaic` | closed when none | 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 eight hold, unnamed otherwise: hostcheck items 1 to 6, read from `/usr/local/lib/mosaic-stack/hostcheck.sh --check`'s own `ok N` lines (item 7 is P10's floor); `sudo -u jarvis /usr/bin/bwrap --version` fails with "Permission denied"; `getent group mosaic` lists no member and no other account has its gid. F5 adds the label of a real role session | the hostcheck lines and each item |
| P10 | a session binds the tester port 7333 while the Console is down | accepted mos-1101 while the port floor is above 732 and 7333 is held by the Console's MainPID | the floor, the owner pid |
| P11 | the boot check runs before the bus | closed when `mosaic-hostcheck.service` is enabled and `systemctl show -p Before` on it names `mosaic-bus@<business>.service` for the args file's business; unnamed otherwise (section 7) | the `Before=` value |

P9 reads hostcheck's output, which root's own script prints, by matching `^ok [1-6] ` lines. It runs `--check` only,
never `--enforce`, so the acceptance can't close bwrap.

Tests, in `packages/cli/tests/case1-accept.test.mjs`. The script takes a test seam, `--root DIR`, for
`/proc`, `/etc`, `/srv/mosaic` and `getent`, and refuses it when run as root, as the hostcheck does. Fake
`systemctl`, `curl`, `ss`, `sudo`, `bwrap`, `sysctl`, `stat`, `claude`, `node` and `hostcheck.sh` on PATH.
Nothing reaches .53 or dragon-lin's live state, and no test reads a real credential. Cases:
- every check passing exits 0, with the counts matching the lines;
- each F check failing alone exits 1 with its line;
- F5 fails on `unpriv_bwrap (enforce)`, `bwrap//&unpriv_bwrap (complain)` and a trailing space;
- F6 fails on a stub `claude.exe` that prints nothing and on a path outside `claude-<pin>`;
- F8 fails on mode 0640, on another owner and on a symlink, and the stand-in credential's value never
  appears in the output;
- F11 fails on one `*.json` under `discord/`;
- each P check in each of its marks: P1 with output that grants nothing but lacks the refusal text
  (unnamed), P9 unnamed once for each item missing, P10 unnamed with the floor at 700 and with 7333 held
  by another pid, P11 unnamed with `Before=` naming another business;
- every P check `unnamed` still exits 0;
- a bad args file refuses with exit 2 before any check;
- every in-instance call goes through `sudo -u mosaic -H env PATH=<node dir>:...`, read from the fake
  `sudo`'s recorded argv;
- the hostcheck fake records `--check` and never `--enforce`.

Mutants, each killed by a named case:
1. F1 passes on a MainPID owned by root.
2. F2 accepts a listener on 0.0.0.0.
3. F3 passes without a reply.
4. F5 matches a substring.
5. F6 skips the `--version` check.
6. F8 skips the owner check.
7. F11 ignores `discord/`.
8. P1 closes on any output.
9. P5 maps to `closed` with `jarvis` in docker.
10. P9 drops the jarvis item.
11. P11 accepts any `Before=`.
12. A P check's `unnamed` sets the exit to 1.

## 6. The .53 window

Nothing in this row touches .53 (sage-1809). Every root step below is ops-01's, over
`ssh [email protected]` with `sudo`, and waits on a Jason decision (mos-1086). The window needs:
- row 226 landed;
- Jason's or Mos's go naming the commit, the bundle's sha256, the node and claude pins, the role set and
  Jason's human id;
- a backup first;
- a STATE report after.

| # | Step | Runs as | Source |
|---|---|---|---|
| W0 | re-read section 0's items; stop on any change and report it | `localadmin`, read only | section 0 |
| W1 | PBS backup of VM 1008 | ops-01 on pve | section 4 |
| W2 | the stack's user and home | root | run sheet 1 |
| W3 | node under `/opt` | root | run sheet 2 |
| W4 | claude under `/opt`, managed settings | root | run sheet 3 |
| W5 | bubblewrap, the profile, the statoverride, the no-new-privs test, the drift stop; `hostcheck.sh --check` exits 0 with item 7 at 1023 | root | run sheet 4 |
| W6 | the port floor (read) | root | run sheet 5 |
| W7 | the bundle, root's checkout at `<sha>`, `mosaic`'s clone | root, then `sudo -u mosaic -H` | run sheet 6 |
| W8 | `case2-install.sh --profile factory` with section 1's inputs: bootstrap, the business file, roles, human id, the copy | `sudo -u mosaic -H env PATH=...` | row 226 |
| W9 | place `claude-session.env` | Jason, or who he names | run sheet 9, section 3 |
| W10 | write `dev.args` (section 1); `case2-unit.sh` from it, which enables both units and starts the bus, then the Console | root | run sheet 7, 10 |
| W11 | `case2-precheck.sh --business mosaic-stack --profile factory` exits 0; `hostcheck.sh --check` exits 0 with item 7 at 732 | `mosaic`; root | row 230, 240 |
| W12 | `stack-release.sh adopt mosaic-stack --units external`, unless row 226's install does it | `sudo -u mosaic -H` | row 232 |
| W13 | `case1-accept.sh` (section 5); the output to Duckworth, then Mos confirms | root | this row |
| W14 | STATE report to Mos | ops-01 | brief |

Install rollback, any step before W13 passes. Stop and disable both units, remove their files and
`dev.args`, then the run sheet's rollbacks for sections 4, 3 and 2 in that order. Removing the user and
home (`userdel -r mosaic`) needs its own Mos go, since it deletes the home. W1's backup covers the rest.

What row 195 needs from row 226. Row 226 is assigned to Gosalyn, briefed and not started, so I'm listing what the window
needs from its design, not designing it:

1. `--profile factory` writes the business file with `bus.tester` on and no `library` key, and the roles
   the go names.
2. It writes no token slot and no persona files (its brief says so).
3. It runs before the unit render, since the copy must exist for `case2-unit.sh --copy`, and it starts no
   unit.
4. Who writes `dev.args`. The C2-13b brief says the install doesn't write unit files, and the run sheet
   says "I2's install writes it". `mosaic` can't write `/etc/mosaic-stack`. So either root's part of I2
   writes it, or ops-01 writes it from section 1 at W10. Either works for row 195. The file's content is
   fixed above.
5. Whether it runs `adopt`. If it doesn't, W12 does.

I sent these to Gosalyn on 2026-10-21 (gosalyn-0085 acknowledged), and the run sheet takes whichever answer 226 gives.

## 7. The business id

.53's business id is `mosaic-stack` (sage-1813). It appears in:
- the args file, twice;
- the bus unit `[email protected]`;
- the business file name;
- `case1-accept.sh --business`.

The landed boot unit `packages/cli/systemd/mosaic-hostcheck.service` has
`[email protected]`. So at boot the drift stop runs before .53's bus, and DEFERRED 240
N1 (da3051ccc: `After=mosaic-hostcheck.service` in `mosaic-bus.service.in`) doesn't bind this row.

If a go ever names another id for .53, the window stops before W8, and the N1 fix lands first as its own
reviewed row. P11 is the check on the host. It reads `unnamed` when the boot unit's `Before=` doesn't
name the installed bus.

DEFERRED 240 N2 (the apt hook goes in before the first `--check`) is a run sheet order note. This row
doesn't edit the run sheet, so it stays deferred.

## 8. Rulings this design rests on

- sage-1451: Q1 system units; Q3 the Console on 127.0.0.1 with `ssh -L`; Q4 a rebase goes to a new branch
  name (#426(a), mos-1086).
- mos-1091, mos-1096, mos-1098, mos-1101: the bwrap path, its profile and P9; the port squat and P10; .53
  runs the PM. Their r6 sections stand, and rows 232, 233, 240 and 230 implemented them.
- mos-1108 (Q187) and mos-1118: the installer rows and the scope cut. mos-1122 (Q189): both units start at
  boot.
- sage-1522: claude under `/opt`, root-owned.
- sage-1809: no step on .53, root steps named and left to ops-01, the business id stated.
- sage-1813: 226 in the after list (rev 2718); the id is `mosaic-stack` and 240 N1 doesn't apply.
- Q185: the factory's model account. Q173: credentials placed by Jason or his named person.

## 9. Manifest (planned)

- `scripts/case1-accept.sh` (new), mode 100755.
- `packages/cli/tests/case1-accept.test.mjs` (new). Section 5's cases and mutants.
- `docs/TOOLS.md`: a `case1-accept.sh` entry with usage, marks and exits, beside the hostcheck entry.
- `agents/darkwing/work/queue-195/BUILD.md` (new): run environment, the mutant results, deviations.
- No change to `case2-*.sh`, `stack-upgrade.sh`, `stack-release.sh`, `case1-hostcheck.sh` or the run
  sheet. Two follow-ups stay out of this row:
  - row 230 r2 N1. The "still on copy" line reads the id from the unit file. On .53 neither the install
    nor `stack-upgrade.sh` reaches that line, since both start from stopped units.
  - DEFERRED 240 N2.
- No change to `~/.mosaic-dev`, `~/.config/mosaic-dev`, dragon-lin's units or `~/.mosaic`.

Gate: Duckworth approves this design, then the manifest; the cli suite and every `scripts/test-*.sh`
green on Sage's gate; section 5's twelve mutants killed in a slot Sage grants.