Use when adding new error messages to React, or seeing "unknown error code" warnings.
npx skills add donghaozhang/video-agent-skill --skill "pr-comments"
Install specific skill from multi-skill repository
# Description
Export, preprocess, and fix GitHub PR review comments. Use when user wants to export PR comments, evaluate code reviews, or fix review feedback from CodeRabbit/Gemini bots.
# SKILL.md
name: pr-comments
description: Export, preprocess, and fix GitHub PR review comments. Use when user wants to export PR comments, evaluate code reviews, or fix review feedback from CodeRabbit/Gemini bots.
argument-hint:
disable-model-invocation: true
allowed-tools: Bash(gh ), Bash(jq ), Bash(mkdir ), Bash(sed ), Read, Edit, Glob, Grep
PR Comments Skill
Export GitHub PR review comments, preprocess for evaluation, and fix or reject each review.
Actions
1. Export: /pr-comments export <owner/repo> <pr-number>
Export all review comments from a PR to individual markdown files.
bash .claude/skills/pr-comments/scripts/export.sh $1 $2 $3
Output: docs/pr-comments/pr-{number}/ with one file per comment.
2. Preprocess: /pr-comments preprocess <input-dir>
Clean exported comments for agentic evaluation (removes <details> blocks, adds evaluation prompt).
bash .claude/skills/pr-comments/scripts/batch-preprocess.sh $1
Output: {input-dir}-tasks/ with cleaned task files.
3. Analyze: /pr-comments analyze <tasks-dir>
Show comments grouped by source file with recommended processing order.
bash .claude/skills/pr-comments/scripts/analyze.sh $1
Output: Table showing which files have multiple comments and line numbers (sorted for bottom-up fixing).
4. Fix: /pr-comments fix <task-file.md>
Evaluate a single PR review comment. Read the source file, determine if valid, then fix or explain.
Follow instructions in review-fix.md.
5. Batch: /pr-comments batch <tasks-dir>
Process all task files in a directory. Groups comments by source file and fixes bottom-up to avoid line number shifts.
Critical: Multiple comments often target the same file. Batch mode:
1. Groups tasks by source file
2. Sorts by line number descending (bottom-up)
3. Fixes all comments for one file before moving to next
Follow instructions in review-batch.md.
6. Resolve: /pr-comments resolve <owner/repo> <pr-number> <comment-id> [task-file]
Resolve a PR review thread on GitHub and move task to completed folder.
bash .claude/skills/pr-comments/scripts/resolve-thread.sh $1 $2 $3 $4
Example:
/pr-comments resolve donghaozhang/qcut 102 2742327370 docs/pr-comments/pr-102-tasks/comment.md
What it does:
1. Resolves the PR conversation thread on GitHub
2. Moves the task file to {tasks-dir}_completed/
When to use:
- After FIXED → Resolve and move to completed
- After ALREADY_FIXED → Resolve and move to completed
- NOT_APPLICABLE → Don't resolve (leave comment explaining why)
Complete Workflow
# Step 1: Export comments from PR
/pr-comments export donghaozhang/qcut 102
# Step 2: Preprocess into task files
/pr-comments preprocess docs/pr-comments/pr-102
# Step 3: Analyze (see file groupings)
/pr-comments analyze docs/pr-comments/pr-102-tasks
# Step 4a: Fix single comment
/pr-comments fix docs/pr-comments/pr-102-tasks/comment.md
# Step 4b: Or batch fix all
/pr-comments batch docs/pr-comments/pr-102-tasks
# Step 5: Resolve thread and move task to completed
/pr-comments resolve donghaozhang/qcut 102 2742327370 docs/pr-comments/pr-102-tasks/comment.md
Output Structure
docs/pr-comments/
├── README.md
├── pr-102/ # Raw exported comments
│ ├── coderabbitai[bot]_file_L42_123.md
│ └── gemini-code-assist[bot]_file_L50_456.md
├── pr-102-tasks/ # Preprocessed for evaluation (pending)
│ └── coderabbitai[bot]_file_L50_456.md
└── pr-102-tasks_completed/ # Resolved tasks
└── coderabbitai[bot]_file_L42_123.md
Supporting Files
- review-fix.md - Single comment evaluation instructions
- review-batch.md - Batch processing instructions
Requirements
- GitHub CLI (
gh) - installed and authenticated jq- JSON processing
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.