review-work
Post-implementation review orchestrator. Spawns 5 parallel background sub-agents and demands all pass.
review-work runs the work back through five lenses in parallel, then aggregates. Every lane must pass for the overall review to pass — it's a quality gate, not a vibe check.
Triggers
"review work", "review my work", "QA my work".
The five lanes
| Lane | Asks |
|---|---|
| Goal verification | Does the result actually match the request? |
| Code quality | Is the code maintainable, idiomatic, and free of slop? |
| Security | Are there vulnerability classes the change introduces? |
| Hands-on QA | Does it work end-to-end when actually run? |
| Context mining | What did the agents miss that the codebase already encodes? |
Five background sub-agents fire in parallel — same shape as ultrawork. The aggregator collects pass/fail per lane and decides the overall verdict.
When to invoke
After an ultrawork session, after a long Atlas execution, after any change you're about to commit but want a second pass on. Drop the trigger phrase and the skill takes over.
Source Notes
Aligned with upstream docs/reference/features.md#built-in-skills.