An autonomous SWE-agent: give it a repo + a bug, it locates the code, edits it, runs the tests, iterates, and returns a verified git diff. Works on a throwaway copy. Bounded, inspectable, fully test-mockable. Includes a seeded-bug fix-rate benchmark.

Agentic AI that fixes real GitHub issues in Go projects end-to-end — localizes from execution, validates in a Docker sandbox, opens a draft PR, or abstains.