| name | review |
| description | Run the complete code review pipeline on recent changes. Launches three review agents in parallel to check code quality, error handling, and simplification opportunities. |
Code Review Pipeline
You are running the automatic code review pipeline for Brain Dump. This pipeline ensures code quality by running three specialized review agents in parallel.
Your Task
-
First, identify what files were recently changed by checking git diff --name-only or looking at the conversation history for Write/Edit tool usage.
-
Launch ALL THREE review agents in PARALLEL using a single message with multiple Task tool calls:
Task 1: pr-review-toolkit:code-reviewer
- Review the recent code changes against project guidelines in CLAUDE.md
- Check for style violations, potential bugs, and adherence to patterns
Task 2: pr-review-toolkit:silent-failure-hunter
- Check for silent failures, inadequate error handling
- Look for empty catch blocks, swallowed errors, missing error messages
Task 3: pr-review-toolkit:code-simplifier
- Analyze for simplification opportunities
- Look for duplicated code, overly complex logic, unnecessary abstractions
-
Wait for all three agents to complete.
-
Summarize the findings in a clear format:
- Critical Issues: Must fix before merging
- Important Issues: Should fix, but not blocking
- Suggestions: Nice to have improvements
-
If there are critical issues, offer to fix them. When the user wants fixes applied:
- Work through findings by severity: CRITICAL first, then HIGH, MEDIUM, LOW
- For large fix sets, suggest committing after each severity level to prevent context loss
- After each fix, verify that
pnpm type-check still passes
- Run
pnpm check after all fixes are applied
-
After completing the review, mark it as done by running:
touch "$CLAUDE_PROJECT_DIR/.claude/.review-completed" 2>/dev/null || touch .claude/.review-completed
Important
- Run all three agents in PARALLEL (single message, multiple Task calls)
- Focus on recently modified files only
- Be concise in your summary - the user has access to the full agent outputs
- Always mark review as completed at the end to prevent re-triggering