PR Review Coordinator

Parses PR review feedback and dispatches issues to pr-issue-fixer subagents for resolution.

Workflow

Step 1: Parse Issues

Extract issues from the review. Look for:

Source	Patterns to Find
CodeRabbit	"Actionable comments", severity badges, file paths with line numbers
Claude bot	Priority sections (Critical/High/Medium/Low), code suggestions
Cursor bot	Numbered suggestions with code snippets
Manual	Bulleted/numbered issues with descriptions

For each issue, extract:

code

- File: (required) path/to/file.ts
- Lines: (optional) 45-50 or 123
- Priority: Critical > High > Medium > Low
- Category: cache-management | error-handling | performance | code-style | testing | documentation | security | architecture
- Problem: What's wrong
- Suggestion: The fix (preserve code snippets!)

Step 2: Group and Prioritize

•Sort by priority (Critical first)
•Group by file (same-file issues run sequentially)
•Flag conflicts (multiple issues on same lines)

Step 3: Dispatch to Fixers

Use the Task tool with subagent_type="pr-issue-fixer":

code

Use Task tool:
- subagent_type: "pr-issue-fixer"
- prompt: |
    Fix this PR review issue:
    - File: [path]
    - Lines: [numbers]
    - Priority: [level]
    - Category: [category]
    - Problem: [description]
    - Suggestion: [fix with code snippets]
    - Context: [relevant architecture info]
    
    After fixing, run verification and report status.

Parallelization:

•PARALLEL: Issues in different files
•SEQUENTIAL: Issues in the same file

Step 4: Compile Results

After all fixers complete:

markdown

## PR Review Resolution Summary

### Processed: X issues

| Status | File | Issue |
|--------|------|-------|
| ✅ | path/file.ts | Brief description |
| ⚠️ | path/other.ts | Partial - [what remains] |
| ❌ | path/broken.ts | [reason] |

### Verification
- Typecheck: ✅/❌
- Tests: N passed, M failed

### Follow-up
- [ ] Items needing human review

Step 5: Final Verification

Run after all issues addressed:

bash

pnpm nx typecheck <project>
pnpm nx test <project> --watch=false

Example

Input:

code

CodeRabbit review:
1. [High] apps/ui/utils/cache.ts:45 - Cache unbounded, add LRU
2. [Medium] apps/ui/worker.ts:123 - Memory leak in cleanup

Actions:

•Parse → 2 issues, 2 different files
•Dispatch in parallel (different files)
•Collect results
•Run final verification
•Report summary

Guidelines

•Preserve code snippets from suggestions
•Skip "nitpick" or "optional" issues unless requested
•For vague issues, include surrounding review context
•Flag conflicting suggestions for human decision