Files

74 lines
6.2 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Row 296 / #1802 — R2 scoped acceptance verified; review pending
Base: e0c6142f3a411ed70e32b2c50d01930e0b0641c8. Branch: lena/row296-r2. Reviewer: Filbert. Gate owner: Sage.
Handoff: c6ab9bdc9 preserved as ad84c1e45, rebased as 4a16ebaa4, including
agents/quackerjack/work/queue-296/BUILD.md. Every changed path is in MANIFEST.sha256,
except the checksum file itself (self-hash is excluded). No product file or pre-existing test changed.
Pre-existing test numstat: empty. BUILD-LOG.md and docs/SESSIONS.md are append-only additions.
Binding ruling: sage-3625 replaces the brief's transient-unit start with plain child Consoles and
explicit test stand-ins. **This candidate does not prove genuine outside-agent ancestry.**
It creates no transient systemd unit and makes no systemd CLI call. R8 remains row 209.
## What is real, and what is a stand-in
- Two configs created by bootstrap --instance hub/factory, two marked data roots, real Console
cli.mjs serve --sso --send, real broker runtime and launcher, real path-bound human-cli and
user-cli wrappers, real access loader, CLI writes/reload, grant filters and revocation.
- One existing HTTPS fakeOidc server, two issuer/client identities. Its generated CA is trusted
only through NODE_EXTRA_CA_CERTS in this test's child environment, never any trust store.
- federation-host.mjs replaces verifyHuman's readProcess argument with actual process fields plus
ppid=1, exactly the test pattern Sage named. This stands in for Jason's shell on setup,
grant/revoke and both Consoles' transports. The actual Console parent/path check remains.
There is no product parameter, environment switch or altered verifier default.
- fakeHarnesses replaces only the pi harness adapter() with a scratch shell invoking the existing
thread-adapter.mjs. The existing webui channel is re-exported unchanged. No network model runs.
- Public HTTPS origins are modelled through HTTP loopback requests with matching Host/Origin;
reverse-proxy TLS deployment is outside this scratch proof.
## Acceptance map
All checks live in `FED R7 scratch two-instance acceptance` in
packages/cli/tests/federation-acceptance.test.mjs.
| Requirement | Assertion/evidence |
| --- | --- |
| Isolated instances | different config paths/data roots/ports; scratch HOME/TMPDIR; two named markers from bootstrap; port 0, forbidden live ports 7330–7332 checked |
| Jason is collaborator, not owner | owner melanie; access user jason-test, home factory, fake sub jason-test-sub; hub /api/config user=true |
| One project, one granted instance | actual access grant CLI, projects=[shared], instances=[assistant]; user project list contains shared only |
| Thread and reply | create thread on shared, POST message, correlated RESULT with in_reply_to=sent.id, from_role=assistant, to_role=user, body contains question |
| Other project refused | private project absent from listing; create on private returns 403 access-refused |
| Board reads refused | /api/board, /api/bus/inbox and /api/bus/tasks return 403 owner-only |
| Factory sees no hub objects | Jason sub refused at factory; factory owner sees factory-only project and no threads; hub cookie gets 401 at factory |
| Revoke takes effect | actual access revoke CLI; existing hub session 401 and re-login 403 |
| Claim and process ended | live run before revoke; end log exitCode=0; no registered sessions; engine and runner pid/startTime gone; readonly claim history has revoke/access-revoked before the process-end wait, latest op=revoke afterward |
| Normal cleanup | EXIT trap terminates only recorded pid/startTime-checked groups; checks no live group member, no recorded TCP/Unix listener, no other scratch Unix listener, removes scratch directory/socket files |
| Killed-driver cleanup | --kill-after-start prints both-Consoles-started marker then SIGKILLs test driver; shell performs the same independent cleanup; expected nonzero run with successful cleanup line |
## Verification commands (executed on sage-4035 CLEAR)
- scripts/test-federation-acceptance.sh — must exit 0, one acceptance test passes, cleanup line present.
- scripts/test-federation-acceptance.sh --kill-after-start — must exit nonzero, print the explicit
interruption marker and `R7 cleanup: no child groups, ports or sockets remain; no units created`.
- python3 agents/lena/work/queue-296/mutants.py M1 — once-only R5 grant projects -> ['*']; must fail
`R7 ungranted project must not appear` and verify cleanup, then restore original source bytes. Python relies on the shell's own run deadline and cleanup; no outer killing timeout.
- python3 agents/lena/work/queue-296/mutants.py M2 — once-only R4 claim end omitted; must fail
`R7 revoke must end the claim as access-revoked` before shutdown, verify cleanup and restore bytes.
- Sage integration gate remains separate: every scripts/test-*.sh and full applicable suites per Sage; not run in this scoped slot.
Status: scoped gate passed: plain 1/1 (exit 0), killed-driver expected exit 1 with verified cleanup, M1/M2 drivers exit 0 with named assertion kills and source restoration; normal whole-file invocation exit 0 with one intentional dedicated-script skip. All cancellations 0. Receipts in run-r2/; counts and commands in BUILD.md.
The script never silently skips a missing tool/sandbox/provider: prerequisites fail the acceptance.
Ordinary CLI suites skip the dedicated test until invoked by its owning script.
Static-review follow-up: filbert-296-first-push-static. C1 removes Python's outer timeout so it
cannot SIGKILL the shell before EXIT cleanup. N2 moves the named claim-history assertion before
the process-end wait. N1's spawn-to-ledger SIGKILL gap is retained and documented in BUILD.md;
--kill-after-start exercises the fully recorded groups. N3 remains a cosmetic fixture limitation.
N4 indentation corrected. N5 scratch-bootstrap config overwrite is explicitly fixture-only.
R2 provenance: source identical to the statically cleared a1321f8f9. Rebase mapping/range-diff
and stable patch IDs are in run-r2/. Only append-log conflict context changed. No unexpected
first-run failure, no old test changed/run, and no broader suite or shared gate claimed.
Manifest coverage is the complete changed-path set over e0c6142f3..candidate, including all
text evidence and this description; only MANIFEST.sha256 itself is excluded from its own hash.