MC-4781

[visibility-only] Queued message recorded for Hermes Luci. MC did not claim the ticket or spawn a runtime.

MC-4781 evidence: task definition is tasks/luci-followup-sweep.md → /usr/bin/python3 /home/lucienne/workspace/scripts/luci_followup_sweep.py. DB row 389683 was a 120.095s scheduler timeout with empty stdout/stderr; surrounding luci-followup-sweep runs completed in ~0.12–0.16s with JSON ok. Manual bounded dry-run against live mc.db completed successfully: EXIT 0, stdout {"dry_run": true, "elapsed_s": 0.065, "backfillable": 0, "backfilled": 0, "overdue_tickets": 0, "overdue_runtimes": 0, "inbox_inserted": 0, "inbox_skipped_dedupe": 0, "status": "ok"}; stderr had JSON progress start/complete. Conclusion: likely transient DB/scheduler contention, but real bug was lack of internal deadline/phase logging/read-only dry-run. Patched and pushed luci-workspace branch kb/mc-4781-luci-followup-sweep-timeout commit 1583ef0b: adds 90s wall timeout (<120s scheduler kill), 5s SQLite busy timeout, nonblocking process lock skip JSON, read-only dry-run, structured phase/error output, and tests. MC-4780 appears to be the duplicate scheduler inbox wrapper; I did not mutate it.

MC-4781 evidence/outcome: inspected scheduler task `tasks/luci-followup-sweep.md` and command `/usr/bin/python3 /home/lucienne/workspace/scripts/luci_followup_sweep.py`. Run 389683 timed out at 120s with empty stdout/stderr; current live dry-run is not reproducing the hang (3 runs completed in ~0.10-0.12s JSON ok; patched-worktree dry-run: `dry_run=true elapsed_s=0.065 overdue_tickets=0 status=ok`). Root cause looks like transient SQLite lock/overlap contention, but there was a real hardening bug: the script had 30s DB waits, no process lock, dry-run could perform schema/backfill writes, and no phase output until final JSON. Patched/pushed luci-workspace branch `kb/mc-4781-luci-followup-sweep-timeout` commit `1583ef0b`: non-blocking `/tmp/luci-followup-sweep.lock`, 90s internal wall timeout, 5s SQLite busy timeout, read-only dry-run, stderr JSON progress, and phase-rich JSON errors. Validation: `py_compile` ok; `uv run --with pytest pytest tests/test_luci_followup_sweep_hardening.py -q` -> `2 passed`; forced SQLite lock exits in 1.006s with `status=error phase=connect_db error=database is locked`. MC-4780 appears to be the duplicate auto-created inbox wrapper; I did not mutate MC-4780 or other human-owned Inbox/backlog tickets.

Controller routed scheduler timeout to internal Kanban card t_d87d37e3 (codexbuilder) in isolated worktree /home/lucienne/workspace/state/kanban-worktrees/mc-4781-luci-followup-sweep-timeout after reclaiming an initial unsafe live-checkout spawn. MC remains visibility-only: no MC pickup/runtime was called. The card will diagnose luci-followup-sweep task run 389683 / inbox row 144327, reproduce with bounded timeout, patch only if needed, and report evidence back here. MC-4780 appears to be the duplicate auto-created inbox wrapper for the same scheduler timeout; left unchanged to avoid inbox cleanup semantics.

Controller gate passed for MC-4781. Independent review initially blocked on dry-run lock-safety; controller fixed it and landed luci-workspace commit 9a765dbd to origin/master and live /home/lucienne/workspace via fast-forward. Verification from live checkout: python3 -m py_compile scripts/luci_followup_sweep.py OK; uv run --with pytest pytest tests/test_luci_followup_sweep_hardening.py -q -> 2 passed; live dry-run against /home/lucienne/workspace/mission-control/mc.db completed status=ok in 0.063s and did not create the lock file. Internal Kanban card t_d87d37e3 marked done. MC-4780 is already auto-closed as recovered. MC remains visibility-only; no MC pickup/runtime endpoints were used.

[visibility-only queue hygiene] Closed stale queued_messages row 1885 because MC-4781 is already terminal/done with controller gate evidence. This only clears delivery residue; it did not launch MC pickup/runtime or mutate unrelated Inbox/backlog rows.