Maintain

---
name: maintain
description: SDLC stage 6 (Maintain) - close the loop - turn incidents, alerts, metric breaches and repeated mistakes into new intent.md files, evals/regression tests and CLAUDE.md or skill updates. Only in repositories where the SDLC is enabled (prompts carry an [SDLC] status line): use when investigating a production incident, failing metric, flaky CI trend, postmortem, or when Claude repeats the same mistake.
---

# Stage 6 - Close the loop

## Incident or metric breach → new intent
1. Gather evidence first: logs, metrics, error traces, deploy history, related PRs and
   their `docs/sdlc/*` artifacts. Quote facts, mark hypotheses as hypotheses.
2. Run `sdlc new <slug> incident` and fill `intent.md`, adding an **Evidence** section
   under *Problem*. *Proposed outcome* states the healthy metric/behaviour.
3. Hand over for `sdlc approve intent`; the fix then follows the normal flow
   (a small fix may use `sdlc skip spec <reason>`).

## Prevent recurrence
- **Regression test / eval:** every fixed incident gets a test or eval that would have
  caught it, wired into CI.
- **Institutional knowledge:** if Claude or the team made the same mistake twice, add a
  one-line rule to `CLAUDE.md` (repo-specific) or to a skill (organisation-wide). Keep
  CLAUDE.md short: replace stale rules instead of appending forever.
- **Gates:** if a problem slipped through a gate, propose a change to `REVIEW.md`,
  `sdlc.config.json` (verify commands, gated commands) or CI.

## Metrics worth watching
First-pass CI success, time from plan approval to PR, rework cycles per PR, review
comments needing human action, defect escape rate, time from alert to `intent.md`.
`.sdlc/audit.log` records approvals, gate denials and verify results for this repo.