MC-4548 — Pool-slot context leak: worker pulled unrelated work (9e22475) into MC-4460 slot

Home Board MC-4548

Pool-slot context leak: worker pulled unrelated work (9e22475) into MC-4460 slot

Factor #2 from MC-4534. An MC-4460 worker committed _disk_pressure_critical fix in luci_operator.py (9e22475) unrelated to its ticket, never pushed. Investigate whether pool res...

State Done Next Action Closed Owner Luci Runtime Closed Age 12d ago

Description

MC-4548

Factor #2 from MC-4534. An MC-4460 worker committed _disk_pressure_critical fix in luci_operator.py (9e22475) unrelated to its ticket, never pushed. Investigate whether pool reset-on-claim or context brief leaks another ticket's task into the worker. Check worktree_pool.py reset + dispatch brief assembly in mc_pickup.py.

Activity

done

Luci is working...

Details — Done · High · Luci ▼

State

Done

Closed

State

Priority

People

Owner (assigned to)

Controller

Timing / Details

Due Date

Snooze

Source luci (luci)

Project Mission Control

Created 12d ago

Updated 12d ago

Advanced / Operator evidence

Routing owner

Routes via

Operator console

Evidence

Ticket is done; runtime is closed. ticket_marked_done profile claude_opus_1m_high · cwd /home/lucienne/workspace/mission-control/.claude/worktrees/pool-1 · uptime 11d 21h · last activity 11d 18h 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.

lucienne 12d ago

Controller triage: valid follow-up from MC-4534 acceptance. Pool-slot context leak caused cross-ticket work contamination. Moving to todo for dispatch.

lucienne 12d ago

Controller: keystone MC-4534 done. Unblocking for investigation.

Luci 12d ago

MC-4548 root-cause: NOT a pool/brief leak — both named mechanisms disproven; 9e22475 was mis-attributed to MC-4460. Findings (evidence-backed): 1. Pool reset-on-claim CLEAN: _reset_slot does reset --hard origin/master + clean -fdx, base keyed on ticket_id (MC-4464 remote base), slot_branch_matches refuses poisoned slots. No cross-ticket carryover. 2. Brief assembly CLEAN: mc_pickup + ticket_runtime build prompts strictly from ticket["id"]'s own desc+comments. Verified live: MC-4460's description is purely ".gitignore audit + schema-version stamp" — zero disk/operator tokens. 3. Real cause: luci_operator.py _launch_operator_dev_loop / _launch_breakglass_dev_loop spawn claude -p --dangerously-skip-permissions with cwd=~/workspace and NO runtime_pool claim. Operator repair fixes commit onto the shared mission-control main checkout. 9e22475 is linear on master (dc5ad66→9e22475→76fa7ba); it only appeared under slot2/mc-MC-4460 via origin/master base advancement during concurrent MC-4460 churn (reflog 09:09). Mis-attribution, not a leak. _disk_pressure_critical was itself the launch trigger, so the false-critical kept launching the loop that fixed it. Deliverable: docs/incident-2026-06-01-pool-slot-context-leak-MC-4548.md (commit f72cd50, pushed slot1/mc-MC-4548). No code-behavior change shipped — correctly gated. Follow-up filed MC-4556: pool-isolate the MC-reachable operator dev-loop (break-glass stays un-isolated by necessity); control-plane recovery change → needs plan+council before shipping.

lucienne 12d ago

Controller close: investigation complete. Root cause = mis-attribution (un-isolated operator dev-loop), not pool/brief leak. Incident doc shipped (f72cd50). No code-behavior change — correctly gated. Follow-up MC-4556 filed for operator pool isolation.