reviewlisted

# Review Read `${CLAUDE_PLUGIN_ROOT}/SKILL.md`, then `${CLAUDE_PLUGIN_ROOT}/references/laws-closing-and-review.md` (L17 to L19), then `${CLAUDE_PLUGIN_ROOT}/docs/CLI.md`'s two sections `sbe review --write --findings-json` and "The adjudication protocol, as a DATA SHAPE (LT-202.B)". This skill applies both rather than restating them: the finding schema, the dedup rules and the blocking rule live in `brothersbe.cli.normalize_review_findings`, and this skill never re-implements them. ## What a review is pointed at Aim matters more than effort here: an independent code review has found a Critical that six adversarial rounds missed, because it was pointed at the contract rather than at execution edges. So review in this order: 1. The **contract**: does the change do what the purpose brief and the ADR said it would, and does it break anything downstream that depends on the old shape. 2. The **design fidelity**: where the implementation and the dossier disagree, one of them is wrong. Name which. 3. The **execution edges**: concurrency, idempotency, transaction boundaries, error paths, retries, partial failure, duplicate delivery. 4. The **evidence**: was the figure checked, was the migration rehearsed forward and back, was the approval real, did the command actually run. ## 1. Mechanical score and gates, before anything a model decides ``` "${CLAUDE_PLUGIN_ROOT}/bin/sbe" review <dossier> ``` Without `--write` this only prints: the scored surface (`sbe_score.py --s