fix(kanban): reject recycled PIDs as prior worker owners by Kyzcreig · Pull Request #956 · ANG-Ventures/hermes-agent

Kyzcreig · 2026-09-24T08:50:55Z

Stacked on #952 for its stuck-card notifier. Verify process start within the claim/spawn event window before refusing a new claim; skip explicitly runtime-bounded old runs. Page continuous prior_worker_still_alive claim refusals after 15 minutes through the existing guard-stuck #alerts lane, with prior PID and diagnostic command. Tests: 211 passed, 1 skipped across second-claim, reclaim, DB and watcher suites; ruff passed. Start-time mutant fails the real-PID claim test. No live board mutation or deployment.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

Kanban card: t_089fcb25 (Argus PASS r5 run 8950)

Kyzcreig · 2026-09-24T08:51:46Z

Upstream surface audit: NousResearch/main and upstream PR NousResearch#121177 do not contain _prior_worker_still_alive, _spawned_owner_alive, or this claim-rejection reason (git grep on both refs returned no hits). Upstream has a separate _pid_recycled guard for reclaim in kanban_db_dispatch.py, so porting this fork-specific claim guard would create a new subsystem rather than fix an existing path. The generic stuck-card notifier being extended here is linked to the upstream port at NousResearch#121177. This PR is intentionally stacked on fork #952; land #952 first, then rebase this PR onto main.

Minimum of #952 (t_7d7ff489) that the prior_worker >15-min page needs, so #956 lands standalone on main: respawn_guard_stuck_tasks + constants, requeue_task + 'requeued' operator kind + kanban requeue CLI (the active_pr clear verb), and the watcher probe/notifier/sender/loop wiring with their tests. #952's dependency_wait resume and PR-intent consumption changes are NOT carried; #952 rebases on top. Verified: focused kanban db/cli/second-claim/reclaim/watchers suites 205 passed, 1 skipped; ruff clean.

…r review skip Argus r2 caveats on #956: C1: the prior_worker_still_alive page hardcoded 'READY card'. The probe now carries the card status and the page says '<STATUS> card' (REVIEW for the review claim door). Dispatcher STUCK log no longer says READY. C2: new test_active_pr_stuck_page_is_ready_lane_only[ready|review] gates the status != 'ready' skip for the active_pr page. Verified: removing the skip -> [review] RED; hardcoding READY in the sender -> test_guard_stuck_sender_routes_to_alerts RED; dropping status from the probe -> both prior_worker parametrizations RED. Focused db/second-claim/reclaim/watchers/cli + stdin-guard tests: 213 passed, 1 skipped; ruff clean; scripts/check_subprocess_stdin.py clean. Upstream ref: NousResearch#121177

Kyzcreig · 2026-09-24T10:36:26Z

Unstacked onto main (t_4a427d98). Head 8ccf5b7, base main. Land order: #956 first, then #952 rebases on top.

Carried from #952, in commit 876e124 only: respawn_guard_stuck_tasks and its constants, requeue_task with the 'requeued' operator kind and the kanban requeue CLI (the clear verb for active_pr), and the watcher plumbing: _guard_stuck_cards, _GuardStuckNotifier, _send_guard_stuck_alert, the guard_stuck arg on _stall_streak_is_bad, and the dispatcher-loop wiring. Tests came with it, including d55f4e7, the corruption-probe call-count fix that the added watcher call requires. NOT carried: #952's dependency_wait resume and its PR-intent consumption changes.

The code Argus verified in r1/r2 is byte-identical to d36f28b: claim_task, _prior_worker_still_alive, _spawned_owner_alive, _real_pid_started_in_claim, respawn_guard_stuck_tasks at the cherry-pick, plus the watcher functions. I checked this with an AST compare.

C1 and C2 are fixed in 8ccf5b7. The page now uses the card's status ('REVIEW card: …'). The active_pr ready-only skip is gated by test_active_pr_stuck_page_is_ready_lane_only, which goes RED if the skip is removed.

Upstream: NousResearch#121177

Verified 170 passed, 1 skipped in targeted kanban suites. Reused-PID mutant fails the real-PID claim test.

Verified focused kanban and watcher suites: 211 passed, 1 skipped. Ruff passed.

Verify ready/review and lane-transition tests: 4 passed; focused DB/reclaim/watchers: 213 passed, 1 skipped; ruff and subprocess stdin guard passed. Review arm failed against ready-only baseline.

…r review skip Argus r2 caveats on #956: C1: the prior_worker_still_alive page hardcoded 'READY card'. The probe now carries the card status and the page says '<STATUS> card' (REVIEW for the review claim door). Dispatcher STUCK log no longer says READY. C2: new test_active_pr_stuck_page_is_ready_lane_only[ready|review] gates the status != 'ready' skip for the active_pr page. Verified: removing the skip -> [review] RED; hardcoding READY in the sender -> test_guard_stuck_sender_routes_to_alerts RED; dropping status from the probe -> both prior_worker parametrizations RED. Focused db/second-claim/reclaim/watchers/cli + stdin-guard tests: 213 passed, 1 skipped; ruff clean; scripts/check_subprocess_stdin.py clean. Upstream ref: NousResearch#121177

Kyzcreig · 2026-09-24T18:21:36Z

🤖 merged-by: apollo · lane: discord · gate: BYPASS: FR paused by Ace ruling 2026-09-22; gate = kanban Argus PASS run 8950 · why: dispatcher: prior-worker PID-reuse check by process start time + >15m claim-refusal page on ready AND review doors (t_089fcb25): Argus PASS-WITH-CAVEATS r5 run 8950 at 9124c85 (rebased onto main after #952, unstacked); CI green

…ecycled PID (t_0ae83825) Rebased onto main after #946/#956/t_21dfa673. The owner window now carries the spawned event's start_token, so termination and liveness paths use the same clock-step-immune identity as the claim guard; legacy rows keep the causal window. enforce_max_runtime passes conn/task_id per main's run-context class guard.

ang-prism · 2026-09-27T13:55:25Z

FleetReview

Review: post-merge · head 7defc243150f · duration 19m 11s
Profile: full recipe · policy: below-size-and-path-gates
Roster: B-assert-ctx → gpt-6-sol (openai), B-state → gpt-6-sol (openai), C-assert-xhigh → claude-code-opus-5-5 (anthropic), F → gpt-6-sol (openai), G → grok-4.6 (xai), L6 → gpt-6-sol (openai)

Post-merge review (fleetreview:post-merge override): this reviewed the merge commit against its first parent — the bytes that already shipped. It is not a pre-merge gate pass.

Confidence: 2/5

Findings

P1 hermes_cli/kanban_db.py:7319 — Unverified owner · agreed: B-assert-ctx,B-state,C-assert-xhigh,F,G (openai, anthropic, xai)
P2 hermes_cli/kanban_db.py:16055 — Wrong PID · agreed: B-assert-ctx,C-assert-xhigh,G (openai, anthropic, xai)
P1 hermes_cli/kanban_db.py:7362 — Live owner bypass · agreed: B-assert-ctx,B-state,L6,C-assert-xhigh,F (openai, anthropic)
P1 hermes_cli/kanban_db.py:7409 — Late fallback event · agreed: B-state,L6 (openai)
P1 hermes_cli/kanban_db.py:16062 — New guard reason shares the existing alert deduplication key · agreed: F (openai)

FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, D=grok-4.6, F=gpt-6-sol · cost: $1.96 · duration: 20m 17s · rounds: 1 · files examined: 6

Kyzcreig force-pushed the fix/kanban-prior-pid-reuse branch from d36f28b to 8ccf5b7 Compare September 24, 2026 10:23

Kyzcreig changed the base branch from fix/t_7d7ff489-respawn-guard-depwait to main September 24, 2026 10:23

Kyzcreig added 4 commits September 24, 2026 08:48

fix(kanban): verify spawned worker process start before refusing claim

3708530

Verified 170 passed, 1 skipped in targeted kanban suites. Reused-PID mutant fails the real-PID claim test.

fix(kanban): page continuous prior-worker claim refusals

a205b32

Verified focused kanban and watcher suites: 211 passed, 1 skipped. Ruff passed.

fix(kanban): page prior-worker refusals from review claim door

a1cf335

Verify ready/review and lane-transition tests: 4 passed; focused DB/reclaim/watchers: 213 passed, 1 skipped; ruff and subprocess stdin guard passed. Review arm failed against ready-only baseline.

Kyzcreig force-pushed the fix/kanban-prior-pid-reuse branch from 8ccf5b7 to 9124c85 Compare September 24, 2026 16:17

Kyzcreig added this pull request to the merge queue Sep 24, 2026

Merged via the queue into main with commit 7defc24 Sep 24, 2026
57 checks passed

Kyzcreig deleted the fix/kanban-prior-pid-reuse branch September 24, 2026 18:29

Kyzcreig added the fleetreview:post-merge Ask FleetReview to review this MERGED pull (merge commit vs first parent) label Sep 24, 2026

Kyzcreig mentioned this pull request Sep 24, 2026

fix(kanban): claim guard sees open prior runs and every spawn of a run #946

Merged

This was referenced Sep 27, 2026

fix(kanban): C4 backfill — close 11 board-gate bypasses (t_027d7fe7) #1358

Merged

fix(c6): fail closed on 28 fail-open / false-success gates (FleetReview backfill, t_6500a97a) #1362

Merged

fix: 7 FleetReview P2 backfill findings (t_f4ef102c) #1410

Merged

Navigation Menu

fix(kanban): reject recycled PIDs as prior worker owners - #956

fix(kanban): reject recycled PIDs as prior worker owners#956
Kyzcreig merged 4 commits into
mainfrom
fix/kanban-prior-pid-reuse

Kyzcreig commented Sep 24, 2026 •

edited

Loading

Uh oh!

Kyzcreig commented Sep 24, 2026

Uh oh!

Kyzcreig commented Sep 24, 2026

Uh oh!

Kyzcreig commented Sep 24, 2026

Uh oh!

Uh oh!

ang-prism Bot commented Sep 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

Kyzcreig commented Sep 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Kyzcreig commented Sep 24, 2026

Uh oh!

Kyzcreig commented Sep 24, 2026

Uh oh!

Kyzcreig commented Sep 24, 2026

Uh oh!

Uh oh!

ang-prism Bot commented Sep 27, 2026

FleetReview

Findings

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Kyzcreig commented Sep 24, 2026 •

edited

Loading