leonvanzyl/agentic-coding-starter-kit/review-pr
- community
Checking that the SKILL.md file is still in the repository.
Review pull requests with complexity-adaptive depth — spawning deep-dive agents for medium and complex PRs. Use this skill when the user says "review this PR", "review PR #123", "check this pull request", "give me a code review", or wants feedback on a PR before merging. Also use when the user says "/review-pr" or pastes a GitHub PR URL. Requires the GitHub CLI (gh).
Install
npx vskill@latest install leonvanzyl/agentic-coding-starter-kit/review-prSKILL.md
Review Pull Request
Review pull requests with a depth that matches their complexity. Simple PRs get a direct review; complex PRs get parallel deep-dive agents analyzing security, performance, and architecture.
Arguments
The user should provide PR number(s) or URL(s). If none provided, ask for them.
Instructions
Step 1: Retrieve PR Details
Use the GitHub CLI to get the full picture:
Step 2: Assess Complexity
Score the PR based on:
- Number of files changed
- Lines added/removed
- Number of commits
- Whether changes touch core/architectural files
Simple (direct review, no agents):
- 5 or fewer files AND 100 or fewer lines AND single author
Medium (1-2 deep-dive agents):
- 6-15 files, OR 100-500 lines, OR 2 contributors
Complex (up to 3 deep-dive agents):
- More than 15 files, OR more than 500 lines, OR more than 2 contributors, OR touches core architecture
Step 3: Analyze
For Simple PRs: review directly — read the diff, check for issues, provide feedback.
For Medium/Complex PRs: spawn deep-dive agents in parallel using the Agent tool with subagent_type: "deep-dive". Each agent gets a focused area:
- Agent 1 — Code Quality: conventions, patterns, maintainability, test coverage
- Agent 2 — Security: input validation, auth checks, injection risks, data exposure
- Agent 3 — Architecture (Complex only): design patterns, separation of concerns, performance implications, backwards compatibility
Each agent receives the PR diff and description and returns a structured review.
Step 4: Vision Alignment
Read the project's README.md and CLAUDE.md to understand the application's purpose. Assess whether the PR aligns with the project's intended direction. Flag significant deviations — not as a blocker, but as a consideration for the reviewer.
Step 5: Safety Assessment
Provide an overall assessment:
- Safe to merge: no significant issues found
- Merge with caution: minor issues that should be noted but aren't blocking
- Needs changes: issues that should be fixed before merging
Include a risk level (low/medium/high) with justification.
Step 6: Report
Structure the review as:
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingcreate-agentic-app/template/.claude/skills/review-pr/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/leonvanzyl/agentic-coding-starter-kit/review-pr)