Autonomous bug-fixing
Rocco takes a failing test and turns it into a fix. Point it at a repo and a test command, and it runs the test, reads the failure, and asks a model to propose a patch. It doesn't trust that patch on faith. It applies the change in a sandbox (local or Docker), then reruns the whole test suite to check nothing else broke. If the fix holds, Rocco hands you a GitHub pull request or a local patch bundle, your choice. There's a CLI for running it from a terminal and a small dashboard for triggering and watching repairs live. The scope right now is narrow on purpose: Rocco expects a test that's already failing rather than writing new ones, and every run is boxed in by the sandbox and a retry limit. It's built with plain Python and split into a few small pieces , orchestrator, sandbox, providers, publisher, as a starting point for figuring out what constrained, trustworthy automated repair could look like.
Category tags:
For Innovators & Creators