All skills
ai-code-review-refactor
ProfessionalRuntime top levelUse review-agent on implementation or repair diffs for hallucinated APIs, unsupported assumptions, unsafe abstractions, and dependency or regression risks. Skip work without a diff or after reviewer edits.
- Group
- Independent review, incident and routing governance
- Routing mode
- evidence-only
- Supported roles
- MainAnalysisTaskReview
- Source
- SKILL.md
Use when
Trigger signals that make this skill the right owner
- implementation diff ready
- repair diff ready for re-review
Do not use when
Anti-triggers — as binding as the triggers above
- no actual diff
- reviewer implemented changed scope
Required inputs
What must be supplied before this skill can decide anything
- acceptance
- boundary summary
- actual diff
- validation evidence
Output contract
What this skill owes the next role
- reviewed and unreviewed scope
- material current-task findings and non-repair adjacent findings
- review-obligation coverage and subsumption
- scoped repair Evidence invalidation and re-review
- residual risk
Escalation signals
When this skill must hand the decision back
- scope, authority, or material risk exceeds the selected task contract
Related expertise
24 Layer 3 candidates may be loaded behind this skill. A single task loads at most three, and only when the current evidence justifies them.
- code-review
- accessibility-inclusive-design
- code-clarity-maintainability
- regression-testing
- domain-object-identification
- implementation-structure-design
- refactoring
- cpp-professional-usage
- rust-professional-usage
- cleanup-deletion-governance
- code-element-professionalism
- design-pattern-selection
- minimal-correct-implementation
- repeat-failure-analysis
- configuration-runtime-policy
- ai-product-extension
- android-platform-extension
- ios-ipados-platform-extension
- windows-platform-extension
- macos-platform-extension
- linux-desktop-platform-extension
- cross-platform-client-extension
- web3-product-extension
- low-level-systems-extension
Targeted references
Each reference carries its own load condition. None of them enter context by default.
| Reference | Load when | Do not load when | Required by |
|---|---|---|---|
| ai-review-pattern-catalog.mdbenchmark-pattern | Findings need pattern-calibrated examples for recurring AI failure modes | The issue is already concrete and no calibration examples are needed | Review |
| checklist.mddecision-checklist | A bounded review needs a compact checklist before approval | Pattern examples or exhaustive gates are required | Review |
| index.mdindex | competing ai code review refactor references require dependency, conflict, or output-fragment selection | the ai code review refactor root or a task-named reference already resolves selection | Review |
| review-output-and-gates.mdtargeted | L5 review needs exhaustive schema, quality gates, handoff routing, or repair/re-review semantics | A compact severity finding list is sufficient | Review |
| solution-optimality.mdtargeted | An AI-generated diff introduces a material algorithm, data-structure, concurrency, cache, abstraction, or measurable resource-use choice | The issue is already a concrete finding or no material implementation choice changed | Review |