review-skills

A plugin that ships code-review workflow skills for AI agents. The skills cover PR scope, pre-review checks, review coverage, finding quality, equivalent defects, disagreements, fix planning, regression risks, and merge-readiness assessment.

The skills are not hooks and do not intercept anything. Their descriptions state when each skill applies; a compatible host can use those descriptions for discovery.

What ships

• skills/pr-scope-slicer/ — decide whether a change set can be reviewed well in one pass; when it cannot, produce an ordered slice plan where each slice is independently reviewable.
• skills/pre-review-self-audit/ — audit your own change against the checks a reviewer will apply, before requesting review, to remove the predictable first review round.
• skills/single-pass-review-completeness/ — structure one review round so it is complete by construction: enumerate dimensions up front, sweep the whole diff per dimension, and declare coverage.
• skills/multi-lens-review/ — walk a non-trivial change through intent, design, implementation, security, adversarial, and verification lenses, then synthesize the findings into one decision with required actions and residual risk.
• skills/review-finding-quality/ — enforce a quality contract on findings (severity, evidence anchor, expected fix, acceptance condition) so each one is actionable and closable in one round.
• skills/equivalence-class-audit/ — turn one confirmed defect into a locked-scope audit for equivalent defects across sibling fields, mirror use sites, bounds, contracts, paths, modes, tests, and docs.
• skills/review-disagreement-resolution/ — resolve a stalled reviewer/author dispute by classifying it as fact, standard, or preference, anchoring it to a source, and applying a decision rule.
• skills/fix-batching-and-root-cause/ — plan a fix batch by clustering findings with a demonstrated shared cause, proposing changes and verification, and labeling the planned fix depth. It does not implement or verify the fixes.
• skills/fix-blast-radius/ — before a fix is pushed, identify what it could newly break and which resolved findings it could regress, with one verification step per risk.
• skills/review-cycle-gatekeeper/ — assess closure quality across review rounds using supplied finding states and verification evidence, audit unresolved threads, and report a pre-merge gate decision.

Skill names

Use these names to select a skill where the host supports explicit invocation:

pr-scope-slicer
pre-review-self-audit
single-pass-review-completeness
multi-lens-review
review-finding-quality
equivalence-class-audit
review-disagreement-resolution
fix-batching-and-root-cause
fix-blast-radius
review-cycle-gatekeeper

Where each skill fits in a review cycle

Scope

• Skill instructions and references only. No hooks or execution interception.
• Actual discovery and explicit invocation syntax depend on the host.