Autonomous Code Review Agent — AI skill for Claude Code

Margin — Autonomous Code Review Agent

A small web app that reviews pasted or uploaded code. Click **Analyze** and it:

1. Checks real syntax — runs an actual language parser (node --check for JavaScript, python3 -m py_compile for Python), not just an AI guess. This catches every syntax error reliably, because it's a real parser doing the check, not a model reasoning about text.
2. Runs the code — actually executes it (via node / python3) with an 8-second timeout, and shows you stdout, stderr, the exit code, and whether it crashed. This surfaces runtime errors (exceptions, crashes, bad exit codes) that only show up when the code actually runs.
3. Reviews it with AI (Google Gemini, two chained calls: scan → synthesize) — finds bugs/quality issues and style/convention issues beyond what the syntax check and run already caught, and produces an overall risk score and summary.

⚠️ **Security note:** step 2 actually executes whatever code you give it, directly on your machine, with your user permissions and no sandboxing. Only run this locally, only with code you trust, and never deploy it somewhere public — an exposed `/api/run` endpoint would let anyone who finds the URL run arbitrary code on your server.

Everything lives flat in this one folder — no `server/`/`public/` split, no build step, and no external frontend framework loaded from a CDN.

code-review-agent/
├── index.js          Express server — API key, syntax check, code execution
├── index.html         )
├── app.js             ) frontend, plain JS, no build step
├── style.css           )
├── package.json
└── .env.example

Your Google API key never reaches the browser. The frontend calls your own server, and the server calls Gemini using a key that only it knows.

---

Get a Google API key

1. Go to aistudio.google.com/app/apikey and sign in with a Google account.
2. Click Create API key, and pic