Run the triage script to find the top cops to fix:
bash1python3 .claude/skills/triage/scripts/triage.py $ARGUMENTS
If no --input is provided, it auto-downloads the latest corpus-results.json from CI via gh run download.
Common invocations:
/triage— top 30 cops by divergence/triage --limit 50— show more/triage --department RSpec— filter to one department/triage --exclude-department Layout— skip complex layout cops/triage --fp-only— only cops with false positives/triage --input /tmp/corpus-report/corpus-results.json— use local file
After identifying targets, use investigate-cop.py to see all FP/FN locations with source context:
bash1python3 scripts/investigate-cop.py Department/CopName --context --fp-only
Then read the cop source in src/cop/<dept>/<cop>.rs and the vendor spec at vendor/rubocop*/spec/ to understand the root cause before fixing.
This skill is read-only. Parallel-agent activity may leave unrelated local modifications in the working tree; do not edit or revert them while triaging.
If triage transitions into code edits, switch to a dedicated git worktree before making changes, then use a fix skill workflow that integrates retained progress back into main before ending the run.