reviewlisted

# review — feature-level audit Runs after `build` (all waves green + verification gate). Isolated subagent work via `agents/reviewer.md`. Contract: [`subagent-fanout.md`](../../references/subagent-fanout.md). Next step: `pact ship`. ## Progress tracking `TodoWrite`: one item per phase, plus one per review pass. ## Phase 0 — schema gate `pact schema --gate`. ## Phase 1 — mode - `--project` -> **project audit**: dispatch a `reviewer` over the whole codebase vs `project.md` + `accepted` DRs + the constitution. Output -> `type: refactor` / `type: chore` spec stubs (`status: draft`) under `docs/specs/`, one per material finding. Blocks nothing. Skip the rest. - default -> **spec review** of the active spec. ## Phase 2 — suite freshness `sha=$(git rev-parse spec/<id>)`. If `pact green check spec/<id> $sha` succeeds and `[review].fresh_suite` is false -> reuse that green; do not re-run the suite. Otherwise run `pact env test` once and `pact green record spec/<id> $sha` on green. ## Phase 3 — dispatch Effort from `--effort` or `[review].effort`: - `quick` -> 1 pass, checks 1–2 only (AC coverage + consistency). - `standard` -> 1 pass, the full checklist. - `deep` -> `--passes` (or `[review].passes`) independent `reviewer` agents, clamped to 2–3 — there are only three focuses to hand out. One focus per pass, in priority order `correctness` / `security` / `architecture`; 2 passes drops `architecture`. `model:` from `[review].model` (or `advanced`). Announce the