mh1301/ai-code-review-agent — reverse-engineered prompt

Build me a simple Python service that can automatically review GitHub pull requests with an AI model.

I want it to run as a small Flask web server with a GitHub webhook endpoint. When a PR is opened or updated, it should fetch the PR diff, send the changed code to a configurable LLM, and post a clear review back on GitHub. The review should include an overall score from 0 to 100, a short summary, inline comments on specific problem lines, and a final approve or request changes decision.

Please include checks for likely bugs, security problems, performance issues, and code style suggestions. Also add a manual API endpoint where I can send owner, repo, and PR number to trigger a review myself, plus a health check endpoint.

Use simple config files or environment friendly setup for the GitHub token and LLM key. Make it easy to run locally with pip install and python main.py, and include a README with setup steps and an example curl request.