sdlc-auditlisted

# sdlc-audit The **artifact** check. Every other gate in LoopSmith reviews a *unit of work*: `sdlc-review` judges one diff, `sdlc-retro` asks what one goal taught, `sdlc-plan-review` holds one plan to the north-star. `sdlc-align` is the only one with a wider lens — and it reads **goals and commits**, the trajectory of what you chose to work on. It never opens the codebase. **`sdlc-align` is Lens B for strategy; this is Lens B for architecture.** A codebase becomes a god-object the way a strategy gets rewritten: through a thousand individually-acceptable changes, each of which passed review on its own merits. Axis 6 catches a smell inside one diff. Nothing catches the smell that took forty diffs to form — that is this skill's job. **Read-only by default.** It produces a report. It files issues only when explicitly asked, and it never edits the north-star, the decisions registry, or any code. ## 1. Ground it on measurement, never on a re-read Run the read-only collector first and reason from its pack: ```bash bash "${CLAUDE_SKILL_DIR}/scripts/audit-collect.sh" --churn-days <N> ``` It emits `{schema, window, degraded[], totals, files[]}` — per source file: `lines`, `churn`, `score` (size × churn), `has_test`, `markers`, `last_commit` — **ranked and capped**. It renders no verdict; that is your job. Honour its `degraded[]` codes: a thin pack is a thin finding, not a manufactured one. **Read the files the pack ranks, not the tree.** That bound is the whole point — an audit