daily-fix: context-headroom check at review-round boundaries
kind: infra#wf-fix#wf-fix-fp:d9763804e883#daily-auto-filed
track:Highlight text on any card (body, plan, or an event) to anchor a comment, or leave a whole-task comment here. Mention @claude to summon a reply.
No comments yet.
Overview / Motivation
Auto-filed by /daily 2026-07-29 (problem sweep; emitting source: miner I-P1 (session 97867df6, #1776)).
Goal
A review round should not start when the session cannot finish it — the death loses the in-flight fix and the successor re-pays it.
Workflow gap
- Bug observed: unverified hypothesis — verify at plan time: the session hit 'Prompt is too long' twice mid-review-fix and the ~40-min successor re-did the fix (miner-read from transcript; not re-probed).
- Why it is a workflow gap: long multi-round sessions predictably exhaust context; the SKILL has autocompact-thrash guidance for subagents but no orchestrator-side headroom check at round boundaries.
- Confidence (emitter): medium
- verified-at-filing: n/a — behavioral gap; the incident evidence is the transcript (97867df6), not grep-verifiable.
Proposed change (refine in planning)
Add a round-boundary duty: before dispatching a new review round, if context is near the ceiling, first land durable state (commit fixes, post the round marker) so a death costs nothing; optionally end the turn and let the tick re-drive.
Scope / surfaces
- Primary target:
.claude/skills/issue/SKILL.md - Grep the workflow surface for the pattern before editing and update every hit.
Constraints / invariants
- Workflow-surface only;
scripts/workflow_lint.py --check-askspasses; ruff clean on touched files.
Provenance
- workflow_fix_target: .claude/skills/issue/SKILL.md
- fingerprint: d9763804e883
Activity