The Problem

In many teams, code reviews are manual, inconsistent, and time‑consuming. The skill code‑review automates the first pass: it pulls the diff, generates a review checklist and comments, and posts them to the issue tracker.

What This Does

skills/engineering/code-review/SKILL.md defines the skill’s behaviour. The companion file skills/engineering/code-review/agents/openai.yaml registers the skill with the OpenAI agent. The skill reads the PR diff via the git command, calls the model with a prompt template, and writes the review to the repository’s PR comment section. The workflow is encapsulated in the SKILL.md file, which references the agents/openai.yaml for the runtime configuration.

How It Is Wired

Entry Point – skills/engineering/code-review/agents/openai.yaml

name: code-review
agent: openai
skill: skills/engineering/code-review/SKILL.md

When the agent is invoked, the OpenAI client loads the SKILL.md prompt, renders the diff (git diff HEAD~1..HEAD), and streams the response back to the caller. Control Flow

  1. openai.yaml triggers the agent.
  2. The agent calls the OpenAI API with the prompt from SKILL.md.
  3. The API returns the review text.
  4. The agent writes the review to the PR comment via the repository’s API.

There are no external calls outside the agent runtime and the GitHub API. The only dependency is the OpenAI client library, which is declared in package.json. The skill’s effect is limited to comment creation; it does not modify code.

Responsibility Map

  • openai.yaml – registration and runtime wiring.
  • SKILL.md – prompt logic and output format.
  • package.json – dependency on openai and @octokit/rest.

The module graph is a simple one‑hop chain: openai.yaml → SKILL.md → API. No cycles or shared hubs exist.

How To Use It

Setup

# Install the plugin that contains the skill
claude plugin install mattpocock-skills@claude-plugins-official

or for Codex / Copilot, use the marketplace add commands shown in the repo’s README.

Configuration Set the OpenAI key in the agent’s environment:

export OPENAI_API_KEY=sk-…

The skill also relies on a GitHub Personal Access Token set in GITHUB_TOKEN for posting comments.

Running It From the repository root, trigger the agent:

# In the agent that supports skills
/skills/code-review

The agent will automatically fetch the latest PR diff and post the review.

Real‑World Use

A continuous‑integration pipeline can invoke /skills/code-review after a successful build. The resulting comments surface automatically in the PR, giving developers immediate feedback without manual review.

Code Health & Issues

  • Tests – 7 test files cover the skill logic (tests/code-review/*.test.js).
  • CI – GitHub Actions workflows (.github/workflows/release.yml, triage-label.yml) run tests on every push.
  • License – MIT, declared in LICENSE.
  • Documentation – SKILL.md and the skill’s README provide usage details.

No static‑analysis issues were reported; the repository contains a package-lock.json, indicating dependency integrity.

The Bottom Line

The code‑review skill is a lightweight, well‑documented automation that integrates cleanly with GitHub and OpenAI. It requires minimal configuration and fits directly into existing CI pipelines. Teams that need rapid, consistent review feedback will benefit, while those with complex review workflows may need additional customization.