[Jammy2211]

Closes #75. Mind prompt: active/repair_queue_automation_filing_gate.md (batch 2026-08-31-pm, member memory-queue-filing-gate, --auto).

The dashboard's per-paper buttons file issues that were never acted on — #69, #71, #72 open since 2026-08-28, all three picked up, all three runs failed. Three distinct faults, three fixes, both files workflow-only:

What changed

1. queue_filing.yml — the gate has been environmentally broken since 2026-08-24. dcd1e2c moved the board theme into PyAutoBrain/board/_theme.py and added the sibling checkout to validate.yml and knowledge_board.yml, but not here — so python -m pytest tests/ -q dies on 30 tests/test_board.py theme-import errors after Claude has already filed the paper, on every run. Added the PyAutoLabs/PyAutoBrain checkout (path: PyAutoBrain, matching validate.yml), placed before the claude-code-action step because Claude's in-prompt step 5 runs the same pytest and cannot fix a missing checkout.
2. queue_actions.yml — the push retry discarded the action on conflict. git pull --rebase origin main || exit 1 lost queue add: A Population Study for Searching Supermassive Binary Black Holes in Active Galactic Nuclei: Continuum Spectral Features and Periodic Variabilities #69's commit to a content conflict against a sibling queue commit. The retry now rebuilds instead of rebasing: fetch, reset --hard origin/main, re-run the action script (idempotent: already-there/already-gone/already-done), re-commit, 3 attempts. A clean tree after re-apply counts as "already landed" only when the re-run returned an idempotent status — an error status exits 1 so the failure step reports it. The close step reports the status that actually landed (steps.push.outputs.status wins over attempt 1's).
3. queue_actions.yml — failures were silent. The close step is implicitly if: success(), so a died push left the issue open with no comment (why queue add: A Population Study for Searching Supermassive Binary Black Holes in Active Galactic Nuclei: Continuum Spectral Features and Periodic Variabilities #69 read as "never swept" rather than "failed"). Added an if: failure() report step, same shape as queue_filing.yml's.

On the prompt's open question (why concurrency: queue-actions didn't serialise #68/#69): the group holds at most one running + one pending run, and near-simultaneous enqueues can race the group check (a third arrival also cancels the pending run). The fix deliberately does not rely on serialisation — the retry is conflict-proof regardless, which is what the prompt requires.

Autonomous-ship gate (five legs, batch launch)

• Tests — make validate + full pytest tests/ -q: 169 passed with the Brain in reach; without it the exact 30 test_board.py failures of the broken CI gate reproduce (red/green basis for fix 1). No downstream library repos (organism repo) — n/a.
• Smoke — n/a: organism/doc repo, no workspace script surface.
• Review — review-faculty surface read (risk flags no-test-changes, config-or-schema — workflow YAML is outside the repo's pytest reach; retry semantics verified instead by a scratch-repo simulation of the exact queue add: A Population Study for Searching Supermassive Binary Black Holes in Active Galactic Nuclei: Continuum Spectral Features and Periodic Variabilities #69 conflict shape, both sibling and our change landing). Verdict CLEAN after resolving the adversary's findings (below). Witness disposition: basis-cited — cannot execute pre-merge (label-triggered workflows run the file from main), so the basis is the local red/green repro plus live-log confirmation that runs 33202478094/33418598499 died at exactly the diagnosed step after the Claude step succeeded.
• Heart — verdict per the shift's human acknowledgement in PyAutoMind/batches/2026-08-31-pm.md (YELLOW 65, acked at dispatch: manifest drift on local checkout origins; CI-status measurement blindness in the web container). No Heart checkout in this container to re-measure — the same blindness the batch record acks; no new reason observed in this task's blast radius.
• Independent adversary — run by a different model against the committed diff, Witness first. Verdict: Witness holds; FINDINGS, no blockers. Finding 1 (retry could exit 0 with the action unapplied when the re-apply errors on the moved main) and finding 2 (close comment reported attempt 1's status) are fixed in b5bc248. Finding 3 is out of this PR's scope and recorded below as a follow-up.

Flagged decision (decide-and-flag, batch launch — the one per this PR): effective autonomy is supervised (= min of the prompt's safe header and the bug work-type cap), which at ship sign-off parks the task until the human's slot. Taken instead: open this PR (the shift's dispatch budgeted this member notify/0 review-minutes and its expected-effect line is "merges into PyAutoMemory; filing automation resumes opening PRs"). Rejected alternative: park awaiting-input on #75, costing the shift. One-command revert: close this PR (gh pr close 76 / the Close button) — no state exists beyond this branch.

Post-merge validation (human checklist — the prompt's close-out)

• Merge, then re-drive queue intake: Strong Lensing Cosmology with Population-level Calibrated Neural Ratio Estimation #71: remove + re-apply the queue-intake label (the trigger is labeled, so it must be re-applied, not just present). Witness: the queue_filing.yml run goes green and opens a filing PR, where every run since 2026-08-24 shows failure.
• Re-drive queue intake: Mapping the Information Geometry of an Unresolved Dark Matter Population using a Differentiable Strong Lensing Simulator #72 the same way (queue-intake off/on) → second filing PR.
• Re-drive queue add: A Population Study for Searching Supermassive Binary Black Holes in Active Galactic Nuclei: Continuum Spectral Features and Periodic Variabilities #69: note queue_actions.yml triggers on issues opened, not labeled, so re-applying interests-add does nothing — instead re-run its failed run 33180301104 (the old push conflict was transient; a re-run uses the original workflow version but should now push cleanly with no concurrent sibling) or tap the dashboard button again for that paper. Expect: paper moves into the reading queue, issue closes with a comment.
• Follow-up (adversary finding 3, out of scope here): knowledge_board.yml and arxiv_refs.yml still carry the discarding git pull --rebase origin main || exit 1 retry and write the same queue files — worth the same conflict-proof shape via $intake.

---

Generated by Claude Code