MC-4502 — Harden Control Room CLI worker launch/liveness ledger

Description

MC-4502

Root cause follow-up from Elmar question about Codex/Claude calls breaking and MC-4494 showing missing tmux. Problems to fix: - Umbrella/interactive websocket rows with no tmux_target can be misread as missing tmux workers. - Background process workers (hermes-bg:proc_*) are sometimes normalized with tmux_session_missing, which is misleading. - Codex dispatch must smoke-test exact model first and never default to unsupported codex-mini under ChatGPT OAuth. - Claude Code print-mode exits such as max_turns/max_budget need clearer classification: partial-useful vs failed vs completed-by-controller. - Completed/recovered tickets should clear stale failure_reason when validator closes them. Acceptance: - Runtime/session UI distinguishes websocket-idle, process-dead, tmux-missing, killed-by-controller, and completed-by-controller. - Watcher does not surface missing-tmux warnings for umbrella/waiting tickets with no active worker expected. - Tests cover MC-4494-style stale idle row and hermes-bg process rows. - Existing completed tickets no longer show stale failure banners after controller recovery.

Activity

done

Luci is working...

Details — Done · High · Luci ▼

State

Done

Waiting: done

State

Priority

People

Owner (assigned to)

Controller

Timing / Details

Due Date

Snooze

Source telegram (human)

Project -

Created 12d ago

Updated 12d ago

Advanced / Operator evidence

Routing owner

Routes via

Operator console

Evidence

Ticket is done; runtime is closed. tmux_session_missing profile claude_opus_1m_high · cwd /home/lucienne/workspace/state/control-room-worktrees/mc-4502-runtime-hardening · uptime 12d 11h · last activity 12d 11h ago MC is visibility-only. Hermes Luci launches and gates work outside MC, then mirrors evidence/status here. Raw console: luci · claude/anthropic/opus[1m]

Workflow

Start Dev Review + QA ▾

Select phases to include:

Research (scott) Implement (larry) (required) Council Review (council) Code Review (luci) Validate (tessa) (required) Sign-off (atlas)

Agents

Review Gates

Decision

WAT routing: choose an agent, review gate, or decision. Buttons use the live runtime when one is attached.

controller 12d ago

Created from Elmar report: Codex/Claude worker calls are too brittle and MC-4494 showed a false missing-tmux state. Initial ledger hygiene already applied; this ticket tracks code-level hardening.

controller 12d ago

Coder gate dispatched: Claude Code Opus/high in isolated worktree mc-4502-runtime-hardening. Codex standalone smoke failed for gpt-5.5 and related models; this ticket will include codex smoke-test hardening.

validator 12d ago

Validator gate complete — MC-4502 landed. Commit: 1bd1c48 Branch: cr/mc-4502-runtime-hardening Master: fast-forwarded and pushed to origin/master. Evidence: - Codex standalone smoke failure reproduced: installed codex-cli rejects required models under ChatGPT OAuth. - Coder produced runtime/liveness hardening; controller recovered max-turns partial work. - Focused validation after final review: `python3 -m pytest tests/test_mc4502_runtime_hardening.py tests/test_ticket_runtime.py tests/test_terminal_state_reconciliation_regressions.py tests/test_runtime_profiles.py tests/test_mc_pickup_larry_runtime.py tests/test_mc4501_comment_author_type_defaults.py tests/test_mc4500_comment_controller_notify.py -q` → 189 passed + 25 subtests. - Compile: `python3 -m py_compile app.py models.py ticket_runtime.py /home/lucienne/workspace/mc_pickup.py` → passed. - Independent review required 4 rounds; final verdict APPROVED. - graphify update completed. - `git merge-base --is-ancestor 1bd1c48 origin/master` → passed. Scope shipped: - Runtime snapshots expose explicit runtime_class. - Waiting/umbrella tickets no longer surface false missing-tmux failure warnings. - tmux_session_missing reaper rows no longer appear as operator-facing provider failures. - Claude max-turns/max-budget is recoverable partial, with immediate harvest-path ticket update. - DONE/REVIEW/QUESTION verdicts mentioning max_turns/max_budget are not misclassified as partial, including decorated prefixes. - Codex model smoke-test helper aborts unsupported model launches before tmux start.