reviewlisted

# Review (phase 6 of 7) Prove the change is done — against the enforcement set, the project rules, and fresh test evidence — before any completion claim. Runs **inline on the main thread** (a subagent cannot spawn subagents). ## Iron Law ``` NO COMPLETION CLAIM WHILE ANY APPLICABLE SKILL, RULE, OR MEMORY ITEM IS UNSATISFIED — AND NONE WITHOUT FRESH REVIEW EVIDENCE ``` If you have not re-run the auditors **this turn**, you cannot say it passes — paraphrases included ("should pass", "looks compliant"). The `Stop` gate blocks turn-end until `review=pass`, and `set-review pass` **requires the `review.md` evidence file**. ## The rigor contract The rules binding the four code-review auditors live in **`references/review-rigor.md`** (think-first · refute on the Spec/Enforcement + Standards axes · evidence per claim · coverage-table output · severity scale). **Cat that file verbatim into each code-review auditor's dispatch prompt** — it is the single source; the auditor bodies do not restate it. `claudehut-test-runner` is exempt (raw test output, no coverage table). (Forgetting to cat it is caught downstream: `set-review pass` refuses any `review.md` whose `✓` rows lack a cited locus, so an uninstructed auditor cannot produce a passing-but-empty review — the cat-instruction is prose, but the gate makes it safe.) ## Flow ```mermaid flowchart TB start([Review phase]) --> lock["SELECT auditor set — criteria before dispatch<br/>(enforcement set + git diff → which reviewers +