AutoFix.sh
AutoFix.sh is an autonomous self-healing deployment platform that monitors GitHub Actions failures, diagnoses build and test errors, generates AI-powered code repairs, validates fixes inside isolated Docker sandboxes, and automatically opens pull requests with verified patches. It pairs a FastAPI repair engine with a real-time TanStack Start DevOps dashboard.
About the Project
AutoFix.sh is an autonomous, AI-powered self-healing CI/CD deployment agent. It watches GitHub Actions failures, diagnoses what broke, generates code repairs, validates the fix in an isolated sandbox, and opens a pull request when the patch is safe.
The project is designed for modern DevOps teams that want deployment failures to become actionable repair workflows instead of manual debugging sessions. It turns failed logs into structured diagnosis, patch generation, verification, and review.
Self-Healing CI/CD,
Validated Before It Ships
The repair flow starts when a GitHub Action fails. A FastAPI webhook handler fetches the raw logs, the parser extracts stack traces, failing files, and line numbers, then an AI classification agent uses GPT-4o to generate a high-confidence unified diff patch.
Before anything is committed, AutoFix.sh spins up a Docker sandbox, clones the repository at the failed commit, applies the patch, and runs the test suite. Only verified repairs move forward to a branch and automated GitHub pull request.
Dashboard and Safety Controls
The frontend is a real-time DevOps console built with TanStack Start, React 19, TypeScript, Tailwind CSS, shadcn/ui, Radix UI, Lucide icons, Vite, and Bun. Server-Sent Events stream repair timelines, command logs, status changes, and code diffs as the agent works.
The backend uses FastAPI, Uvicorn, Docker SDK, OpenAI’s Python API, PyGithub, Structlog, and an asynchronous in-memory event bus. Feature gates control confidence thresholds, sandbox execution, and dry-run commit behavior so automation stays observable and controlled.
Why It Matters
CI failures often slow teams down because the error, diagnosis, fix, and verification are scattered across logs, local environments, and code review. AutoFix.sh compresses that loop into a repeatable repair pipeline.
From Broken Build to Verified PR
The project shows how AI can be applied responsibly in developer tooling: not by blindly changing code, but by proposing fixes, validating them in isolation, and presenting verified patches for review.
