IrredenEngine: Skill for Claude Code

review-fleet-feedback

The deterministic work lives in scripts/fleet/review-fleet-feedback; this skill runs it, reads its JSON, and separates two kinds of decisions:

• Mechanical transitions (pending_mutations in the digest — merged-PR flips, 7-day auto-closures, recurrence flags): applied without asking; the script detected them from GitHub state.
• Human-judgment decisions (which new proposals to log, whether to drop a proposed fix): asked.

Both go back through the script's apply. Signature parsing lives in the script's SIGNATURES table, never in this skill body.

State, all under ~/.fleet/feedback/ and never committed: <role>.md (source entries, read-only here), .last-reviewed (ISO watermark), .fix-log.jsonl (one JSON object per line; tolerates manual edits). If the directory is missing or empty, report "no fleet feedback yet" and exit.

Flow

1. Digest. scripts/fleet/review-fleet-feedback digest > /tmp/fleet-digest.json. Fields: clusters (count ≥ 2 by signature, with examples and all), singletons (never proposable), recurring (applied fixes with entries after applied_at), still_proposed, closed_this_run (applied fixes quiet ≥ 7 days), fresh_proposable (clusters with no open fix-log row), pending_mutations, fix_log_counts. total_entries == 0 → skip to step 5 (still print the open fix-log and bump — the contract is "seen", not "fixed").
2. Proposals. One paragraph per fresh_proposable cluster naming the concrete artifact to change (a file path or script) and any cross-refs. A proposal that cannot name an artifact says so and tags the cluster needs-investigation. One occurrence is noise; two is a pattern.
3. Present, most urgent first:

   RECURRING (previous fixes didn't close the loop):
     ! fix-NNN (<signature>) — <N> occurrences since applied <date>
   
   STILL PROPOSED (not yet applied, still recurring):
     · fix-NNN (<signature>) — <N> new occurrences since proposed <date>
   
   NEW PROPOSALS:
     1. [Nx] <signature>
        <proposal text>
   
   CLOSED THIS RUN (no recurrence in ≥7d):
     ✓ fix-NNN (<signature>) — closed after <N>d clean
   
   Open fix-log: <P> proposed, <A> applied, <R> recurring, <C> closed.
   

   List 3–5 interesting singletons in a collapsed block at most.
4. Reconcile. Mechanical transitions go into the decisions JSON unconditionally. Ask only for: ≤ 4 new proposals → one AskUserQuestion with each as a multi-select option; > 4 → one bulk question (log all / top N / none / let me pick); manual drops and manual applied-confirms only when the human raises them or a still_proposed row is ≥ 30 days old (batch still-proposed / drop when > 3). recurring fixes need a fresh proposal, not a status flip — say so.
5. Apply and bump. Write the decisions JSON (apply --schema prints the schema, including the pending_mutations), then:

   scripts/fleet/review-fleet-feedback apply --decisions /tmp/fleet-decisions.json
   scripts/fleet/review-fleet-feedback bump
   

   If anything errored before this step, do not bump — a stale marker re-surfaces the entries next run.
6. (Optional) file tasks. To hand proposals to the human:approved → fleet-queue-ingest → worker pipeline:

   scripts/fleet/review-fleet-feedback file-tasks --dry-run
   scripts/fleet/review-fleet-feedback file-tasks [--ids fix-001,fix-003]
   

   Each proposed row without a tracking_issue becomes a human:approved issue quoting the proposal and three representative entries; the URL is written back to the row, and a merged Closes #N PR flips it to applied on the next digest. Confirm with the human first — these are public issues; default to --dry-run.
7. Summary. Reviewed N entries since <date> (C clusters, S singletons). M new proposals logged, K previously proposed resolved, L auto-closed. Marker bumped to <ts> UTC. Open fix-log: <counts>.