| name | review-process |
| description | Use when an agent must review code changes. Defines the universal review sequence โ identify files, read context, check against domain criteria, score, filter. Compose with a domain skill and review-output-schema. |
Review Process
Base review protocol for all code reviewers. Every review follows this sequence.
Steps
Step 1: Identify Changed Files
- If given a git range:
git diff <range> --name-only filtered to relevant extensions
- If given file paths: use those directly
- If given a feature description: search for recently modified files
Step 2: Read Full Context
For each changed file:
- Read the complete file โ not just the diff
- Read any project headers it includes (not engine/framework headers) โ you need to see the types being used
- If a changed function takes or returns a pointer/reference,
Grep for that symbol to understand its lifecycle across
call sites
Adapt depth to your mandate: style review may be self-contained per file; safety and correctness review require
cross-file tracing.
Step 3: Check Against Domain Criteria
Systematically check every rule from your loaded domain skill(s). For each file, for each rule category โ be thorough
but precise. Only flag things that are clearly violations, not judgment calls.
Step 4: Score and Filter
Rate every potential finding on a confidence scale. Apply the scoring thresholds and output format from your output
schema. Only report issues you can substantiate with specific code evidence.