// Run a swarm-like PR review using parallel exploration, independent reviewer validation, and critic challenge. Use for deep pull request review with low false-positive tolerance.
| name | swarm-pr-review |
| description | Run a swarm-like PR review using parallel exploration, independent reviewer validation, and critic challenge. Use for deep pull request review with low false-positive tolerance. |
| disable-model-invocation | true |
Use this skill when reviewing a PR, branch diff, staged diff, or recent commit with maximum quality.
Use this layered workflow:
This is intentionally not a full-depth pass on every minor issue. It is a speed-preserving, quality-maximizing review ladder. Parallel breadth stays wide. Deep validation is concentrated where bugs are expensive.
Determine review scope using this priority:
Launch in parallel where scopes are disjoint:
Explorer lanes should optimize for recall and speed. They should produce candidate findings with exact evidence, not final conclusions.
Validate every candidate finding that is:
Reviewer must classify each validated candidate as:
Reviewer should be hyper-critical and suspicious. Default to disbelief until the issue is actually supported by code evidence. If a mitigating runtime control may invalidate the claim, check that before confirming the finding. Lower-risk suggestions can remain lightweight if they are clearly non-blocking and strongly evidenced.
Use critic only after reviewer validation. Critic reviews small batches of reviewer-confirmed findings and challenges:
Produce:
Do not let speed degrade validation quality.
