| name | review |
| description | Pressure-test a proposed architecture, issue, PR intent, or inferred design. Writes detailed findings to .context/review.md and returns LOCKED or NOT LOCKED with blockers only. |
/review
Read the shared contracts before output: ../_shared/operator-output.md, ../_shared/attention-budget.md, ../_shared/evidence-record.md, ../_shared/state.md, ../_shared/cognitive-load.md, ../_shared/collaboration.md.
Purpose
Catch design errors before they become implementation cost.
Inputs
.context/architecture.md, a GitHub issue, PR body, user proposal, or inferred design from current code. Missing architecture artifacts are not blockers; reconstruct the design from available context and label inferred claims.
Reads
Architecture evidence, repo code/tests, dependency docs/source when referenced, git history and prior learnings when relevant.
Writes
.context/review.md
.context/session-state.md
Process
- Identify the design under review and what evidence supports it.
- Check invariants, source of truth, boundaries, effects/errors, state lifecycle, observability, migration, and local incentives.
- Classify findings as
blocking, important, or minor.
- Put detailed findings, diffs, and reality-check notes in
.context/review.md.
- Decide
LOCKED only when no blocking issue remains and important issues are fixed or accepted as explicit tradeoffs.
- Update session state.
Operator Output
Return LOCKED or NOT LOCKED, up to three evidence bullets, the highest remaining risk, the evidence path, and the next natural action. Do not restate the full architecture in chat.
Stop Conditions
Stop when the lock decision is clear, or when a missing product/risk decision prevents judging the design.