[Cuarroc]

Dieser PR fügt einen Test hinzu, der auf einer echten Datenbank beweist, dass "resume" in v1.5.0 geschlossen bleibt. Ohne API-Token kommt 401, mit API-Token und fehlendem, falschem oder gültigem Verdict-Token immer 409. Zustand, Journal, Claims, Runs und Starts bleiben unverändert, auch nach dem Neuöffnen der Datenbank; am Verhalten ändert sich nichts.

Report

• Flake fix: observedAt normalised, run 37269346495 (commit 1e3843e; without_clock strips wall-clock observedAt from the runtime-limits snapshot; prepush lane exit 0, log /tmp/prepush-rc.log).

What changed (2 files, +132/-2): new test api::tests::resume_closed_tests::resume_remains_closed_for_every_verdict_and_preserves_runtime_state (src-tauri/src/api/tests/resume_closed_tests.rs), plus a test-helper change in api.rs: the FakeBackend::control_continuous delegates to native_store when set (otherwise the old canned error). No production code touched. Characterization of matrix row 3, Option A (resume stays closed).

The test covers: 401 without API token (also with only a verdict token); 409 with absent / wrong / current verdict token and the fail-closed message; direct Store::control_continuous(.., "resume") returns the fail-closed error; fingerprint of control row, journal, claims, runs and launch rows plus context cursor and continuousScheduler/launchIntent (both false) unchanged; same after reopening the database and booting a new server.

Evidence

• Base: origin/main 4419432 (PR feat(api): audit planning refusals and writes (M4-R19-08) #444 merged, api lane free). Head f2ba7e4.
• cargo test --bin projecta resume_remains_closed -> exit 0
• bash scripts/ci/gates.sh lane prepush at f2ba7e4 -> exit 0 (log /tmp/row3-prepush.log on the server)
• bash scripts/ci/red-first.sh --plan -> count=1, specs=1
• Red-first: the test does not exist at the merge base (missing = red); trailer Test-First: <path>::<name>. It is a characterization test, so it is green by construction on the unchanged behaviour.

Architecture drift: no new findings.

NICHT ABGEDECKT: #[cfg(windows)] tests and the Windows half (runs in the merge queue); GUI; real main.rs ApiBackend (the test uses the store through the fake's delegation, the route and store code are the production ones); no real worker/launch process.

Review: done, see disposition.

Review disposition

Tier A (api seam, test helper). Reviewers: kimi-k3:cloud and glm-5.2:cloud (Ollama, via scripts/review/run-local.sh); the author family is Claude. Reviewed candidate f2ba7e4; fixes in 5a40bd8. Driver cross-checked every finding against the code.

ID	Reviewer	Severity	Where	Decision
F1	kimi-k3	medium	api.rs:3253-3258, resume_closed_tests.rs:6	Confirmed: fake fallback text equalled the store's, so the route proof did not show the store in the path. Accepted, 5a40bd8: fallback now fake backend without a store: resume is fail-closed (still 409), so assert_closed fails when the store leaves the path.
F2	kimi-k3	medium	api.rs:3253-3258	Rejected: no other test reaches control_continuous through a store-backed fake (planning_access_tests.rs:254 returns 403 before the backend; api.rs:4129 uses a store-less fake). Production main.rs:2923 already delegates the same way. Non-resume actions are out of this package (resume-only characterization).
F3	kimi-k3	low	resume_closed_tests.rs:81-83	Rejected: test-only seed, deterministic defaults() policy, no apostrophes; a migration that changes columns should fail loudly.
F4	kimi-k3	low	resume_closed_tests.rs:9-22	Half accepted, 5a40bd8: continuous_goals and continuous_root_policies added to the state fingerprint. NULL concat refuted: continuous_events.detail is NOT NULL (store/continuous.rs:170).
F5	kimi-k3	low	resume_closed_tests.rs:1-2	Accepted, 5a40bd8: module doc narrowed to a paused project and verdict-token variants.
F6	glm-5.2	low	api.rs:3247	Accepted, 5a40bd8: comment on the sync block_on constraint (same pattern as the neighbouring fakes).

Verdicts: kimi-k3 approve with conditions (F1 addressed, F2 rejected with reason); glm-5.2 approve. Architecture drift: none (existing fake-delegation pattern reused).

Evidence at 5a40bd8: cargo test --bin projecta resume_remains_closed -> exit 0; pre-push hook (gates.sh lane prepush) -> exit 0, push verified with git ls-remote; bash scripts/ci/red-first.sh --plan -> exit 0, count=2.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[Cuarroc]

@Mergifyio queue

[mergify]

Merge Queue Status

• ✅ Entered queue — 2026-10-05 05:47 UTC · Rule: default · triggered by @Cuarroc with the @mergifyio queue command
• ❌ Checks failed · on draft merge queue: checking #453 on main (de0819b) #458
• 🚫 Left the queue — 2026-10-05 05:55 UTC · at 5a40bd80479a67f456b45f95e4d30e089fb308a5

This pull request spent 7 minutes 42 seconds in the queue, including 7 minutes 27 seconds running CI.

Reason

The merge conditions cannot be satisfied due to failing checks

• gates (linux)

Failing checks:

• ❌ gates (linux) (job log)

Hint

You may have to fix your CI before adding the pull request to the queue again.
If you update this pull request, to fix the CI, it will automatically be requeued once the queue conditions match again.
If you think this was a flaky issue instead, you can requeue the pull request, without updating it, by posting a @mergifyio queue comment.

Waiting for

• check-success = gates (linux)
• check-success = gates (windows)

All conditions

• check-success = gates (linux)
• check-success = gates (windows)
• check-success = red-first
• label != do-not-merge
  • test(api): prove resume stays closed on a real store (M4 row 3) #453

Requeued — the merge queue status continues in this comment ↓.

[Cuarroc]

@Mergifyio queue

[mergify]

Merge Queue Status

• ✅ Entered queue — 2026-10-05 06:29 UTC · Rule: default · triggered by @Cuarroc with the @mergifyio queue command
• ✅ Checks passed · on draft merge queue: checking #453 on main (6d5b89a) #464
• ✅ Merged — 2026-10-05 06:41 UTC · at 09d6c078ae7be8cb9e21154ba7b5622787951332 · merge

This pull request spent 12 minutes 6 seconds in the queue, including 11 minutes 49 seconds running CI.

Required conditions to merge

• check-success = gates (linux)
• check-success = gates (windows)
• check-success = red-first
• label != do-not-merge
  • test(api): prove resume stays closed on a real store (M4 row 3) #453