/sg-improve — Self-Improving Feedback Loop

After an audit or visual-test session, extract what went well and what didn't, then feed those insights back into ShipGuard so the next run is better.

Recommended model: Sonnet 4.6. Mechanical retrospective work — use /model sonnet to save Opus quota.

Two outputs:

1. Local learnings (.shipguard/learnings.yaml) — project-specific knowledge, read automatically by the next /sg-code-audit
2. GitHub issue (on {target_repo}, Phase 5) — generic improvements for all ShipGuard users

Phases: 0 snapshot · 1 collect data · 2 extract signals · 3 classify · 4 learnings + 4b mistakes · 5 GitHub (sanitized) · 6 summary.

Invocations