siiddhantt/buildmedic

BuildMedic is an AI assistant that helps developers understand and fix failing GitHub pull requests. When a build breaks on GitHub, it's usually the start of 10-20 minutes of manual detective work, opening logs, finding the error, tracing it back to code, and deciding whether the fix is in your code, dependencies, configuration, or infrastructure. BuildMedic automates that first diagnostic pass, turning a mess of logs and metadata into a clear summary with a proposed fix (or explanation for why code isn't the issue). The tool works through three AI agents that each play a specific role. The Triage Agent digs through the pull request, changed files, failed tests, and logs to pinpoint what broke and why. The Patch Agent then decides whether a code change will fix it, and if so, generates the exact diff you'd need. The Review Agent independently validates both the diagnosis and the proposed fix before presenting the results to you. Every step gets logged and saved locally, so you can review the AI's reasoning, reload old analyses, or delete them. BuildMedic is intentionally read-only right now: the AI investigates and proposes, but never actually pushes changes or modifies your repository. This is a deliberate safety choice, in a future production version, those capabilities could be added behind human approval gates and policy checks, but for now the focus is on getting the diagnosis right. You'd use this if you're tired of being the debugging bottleneck on your team. Instead of developers spending their first 10-20 minutes per failure reading cryptic logs, BuildMedic gives them an auditable summary backed by evidence. It's built on top of GitAgent (a framework for AI agents in software workflows) and pulls structured data from GitHub's API, so it understands the full context of your PR and its failures. The whole thing runs locally, just point it at a failing PR URL, watch the agents work in real time, and review their conclusion.