Please execute the following steps in order.
Step 1: Interpretation of arguments
Interpret $ARGUMENTS according to the following rules:
- First argument: review target (if omitted,
diff= current unstaged changes + untracked files) - Second argument: reviewer name (if omitted, run all reviewers in parallel)
How to specify the review target
- No specification /
diff:git diff+git ls-files --others --exclude-standardto also retrieve new files staged:git diff --cachedbranch:git diff origin/main...HEADPR #123orpr 123:gh pr diff 123- File path: use the contents of that file as the review target
Available reviewer names
code- Quality, security, and performance review by the home-code-reviewer agentsimplify- Readability, consistency, and maintainability review by the home-simplify-reviewer agent (no fixes, feedback only)codex- Review from additional perspectives by the home-codex-reviewer agentrust- Rust-specific review by the home-rust-code-review skill (only for Rust code)design- Design review by the home-design-review skill
If the reviewer name does not match any of the above, treat it as an error and inform the user of the available reviewer names.
Step 2: Review execution
Each reviewerdoes not make fixes, but only reports review feedback.
- If a reviewer name is specified: launch that reviewer's agent, pass the diff information of the review target, and execute the code review
- If the reviewer name is omitted: determine the language of the target code and launch the appropriate reviewerssimultaneously in parallelto execute the review
- If Rust code is included: run
code,simplify,codex,rustin parallel - Otherwise: run
code,simplify,codexin parallel
- If Rust code is included: run
Have each reviewer number their feedback sequentially (e.g., code #1, simplify #1, rust #1). This is used for tracking in Step 3.
Step 3: Fixing review feedback
Once all reviews are complete, execute the /home-fix-review-comments skill to address the review feedback.
This skill does the following:
- Critically evaluate the validity of each item of feedback (not all reviewer feedback is necessarily correct)
- Implement fixes only for feedback judged to be valid
- Output a summary with reviewer identifiers (both the feedback addressed and the feedback not addressed)