Loop Triage Skill — aflare
Overview
Run the daily triage loop for the aflare repository. Scan CI failures, open issues, pull requests, and recent commits. Compare against the current STATE.md and update it with prioritized findings. This is L1 report-only mode — do not auto-fix anything.
Prerequisites
- Read access to the repository
STATE.mdexists in the local repo root (gitignored — loop state is not committed)loop-run-log.mdexists in the local repo root (gitignored)- GitHub API access or local git repo
Instructions
Step 1: Read current state
Read STATE.md to understand:
- Current high-priority items
- Watch list
- Recent noise / ignore rules
- Last run timestamp
Step 2: Scan signals
Collect data from these sources:
- CI Status — Check main branch and recent PRs for failing checks
- Open Issues — Scan all open issues, prioritize by:
- Security-related (label: security)
- Bug reports
- Feature requests
- Open PRs — Check for stale PRs (>3 days without activity)
- Recent Commits — Last 24h of commits, note any reverts or fixes
- Dependabot / Security Alerts — Check for dependency warnings
Step 3: Triage and prioritize
Classify each finding:
| Priority | Criteria |
|---|---|
| High | Main CI red, security issue, data loss, crash, broken core flow |
| Medium | Stale PR, non-critical bug, performance issue, documentation gap |
| Low / Watch | Minor cosmetic, nice-to-have, unclear if real issue |
| Noise | Bot PRs, expected failures, already-known items |
Step 4: Update STATE.md
Update STATE.md with:
- New
Last runtimestamp - New or updated items in High Priority section
- Updated Watch List
- Prune items that are resolved (closed issues, merged PRs, fixed CI)
- Add Post-Run Critique entry
Step 5: Append to run log
Append a row to loop-run-log.md with:
- Date, pattern (daily-triage), level (L1)
- Duration, tokens (if tracked)
- Items found count
- Action taken ("report-only")
- Notes
Step 6: Post-run critique
Add to the critique section:
- One thing that was noise / false positive
- One adjustment to try next run
- Any items that need human discussion
Output
Updated STATE.md and appended loop-run-log.md. No code changes in L1 mode.
Examples
Typical run:
Run loop-triage skill. Read STATE.md. Check CI on main — green. Scan 1 open issue (#6 docs gap). 0 open PRs. No CI failures. Update STATE.md last run timestamp. Append to loop-run-log.md. No new high-priority items. Critique: quiet day; next run add commit scan.
CI failure run:
Run loop-triage. CI red on main — build failure in gofmt check. 2 open issues. Open worktree suggestion for gofmt fix. Update STATE.md with high-priority CI item. Mark as "waiting on human". Append run log. Critique: gofmt failures are recurring; consider adding pre-commit hook.
Resources
Loop state files are maintained locally by the loop operator (gitignored, not committed):
STATE.md— State file to updateLOOP.md— Loop configurationloop-run-log.md— Run logloop-budget.md— Budget limits- Loop Engineering — Daily Triage Pattern