daily-fix: judge-monitor reads are digest-grain, never full
kind: infra#wf-fix#wf-fix-fp:ee760e4eb9a3#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 F-P5 (session with #1776)).
Goal
Monitoring a judge run must not page raw judge rationales (which quote the judged harmful/sensitive content) into orchestrator context.
Workflow gap
- Bug observed: unverified hypothesis — verify at plan time: the refusal trigger was the cat'd raw judge reasoning (temporal correlation in-transcript; content trigger not isolated). Observed: one refused turn, recovered same second by the tick.
- Why it is a workflow gap: trigger-dense-review.md covers run-failure text ingest and review targets; judge-output monitoring reads are the uncovered sibling surface.
- Confidence (emitter): medium
- verified-at-filing: n/a — reason: refusal causation is not grep-verifiable; the rule-surface gap is (planner greps trigger-dense-review.md for judge-monitor guidance).
Proposed change (refine in planning)
Add a judge-monitor clause: digest-grain reads (counts, verdict lines, error classes via grep/jq) for any judge output file; cherry-pick single rows by offset only when needed.
Scope / surfaces
- Primary target:
.claude/rules/trigger-dense-review.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/rules/trigger-dense-review.md
- fingerprint: ee760e4eb9a3
Activity