microboxlabs CodeRabbitAI PR Review and Fix
Review the current branch's PR for CodeRabbitAI inline comments and fix all valid issues.
Prerequisites
ghCLI must be authenticated (gh auth status). No additional tokens required.
Workflow
1. Fetch CodeRabbitAI comments
Auto-detect the PR number, then fetch all inline review comments from the bot:
PR_NUMBER=$(gh pr view --json number -q '.number')
gh api "repos/{owner}/{repo}/pulls/${PR_NUMBER}/comments" \
--paginate \
--jq '[.[] | select(.user.login == "coderabbitai[bot]") | {id, path, line, start_line, body, diff_hunk, created_at, pull_request_review_id}]'
If gh pr view fails (no PR for this branch), ask the user for the PR number or use -p <number>.
If no comments are returned, report "No CodeRabbitAI comments found on this PR" and stop.
2. Parse each comment
Each comment body follows a structured format. Extract these fields:
| Field | How to extract |
|---|---|
| Severity | First line: _🟠 Major_ or _🟡 Minor_ |
| Title | First **...** bold text after severity |
| Explanation | Text between title and first <details> block |
| File | path field from API response |
| Line | line field (end line); start_line if range |
| Suggested fix | Content inside <details><summary>Suggested fix</summary> |
| AI agent prompt | Content inside <details><summary>🤖 Prompt for AI Agents</summary> |
The AI agent prompt is specifically designed for LLM consumption and is the primary guide for fixes.
For parsing details, see reference.md.
3. Group by file and plan fixes
- Group parsed comments by File path.
- Within each file, order by Line number (ascending) so edits applied top-to-bottom keep line numbers valid.
- Read each referenced file to prepare for validation.
4. Validate and apply fixes
For each comment, in file-then-line order:
- Read the actual code at the referenced file and line range.
- Evaluate validity:
- Is the code actually doing what CodeRabbit claims?
- Does the suggestion align with existing project patterns?
- Is this a meaningful improvement or just a style preference?
- Would the fix introduce regressions?
- Has the code already been changed since the comment was posted?
- If valid: Apply the fix using the AI agent prompt as primary guide, cross-referencing the suggested fix diff. Make minimal changes. Preserve existing style.
- If invalid: Skip it. Record the comment title and a one-line reason (e.g., "CodeRabbit misread context", "style preference conflicting with project convention", "code already fixed").
After editing a file, re-read affected regions to ensure no new issues (broken imports, syntax errors).
5. Verify
- Type-check changed files:
npx tsc --noEmit - Run tests for affected modules if available (e.g.,
npx vitest run <path>)
6. Report summary
After all comments are processed, report:
- Fixed (count): Comment title and file, one line each.
- Skipped (count): Comment title, file, and one-line reason.
- Reminder to review changes and run the full test suite.
Severity guidance
| Marker | Level | Action |
|---|---|---|
| 🟠 Major | Likely bug, security, or correctness issue | Prioritize fixing |
| 🟡 Minor | Code smell, missing edge case, improvement | Validate carefully — higher false-positive rate |
Important guidelines
- Never blindly apply all suggestions. Always validate against the actual code.
- Prefer the AI agent prompt over the suggested diff when they conflict.
- One logical edit per fix. Do not combine multiple fixes into a single edit.
- Preserve project style. Do not change formatting or patterns beyond what the fix requires.
- Watch for stale comments. If the code has changed since the comment, skip with a note.
For the full comment schema and edge cases, see reference.md.