DevFix 🛠️

The dreaded "It works on my machine" syndrome. When a developer clones a repository and the build fails due to missing dependencies, cryptic errors, or lockfile mismatches, they spend hours Googling stack traces.

Existing AI tools like Copilot fail here because they rely purely on text generation, guessing the fix without proving it works. Giving an AI access to run arbitrary bash commands on your host laptop to test its fix is incredibly dangerous.

Built DevFix, an open-source agentic workflow available via NPM that detects, diagnoses, and fixes broken environments autonomously.

• Self-Verifying Deterministic Layer: DevFix runs your test suite (npm run build). The agent only completes its task when the environment proves the fix works.
• Isolated Sandboxing: All AI execution runs inside a secure, ephemeral Docker container to protect the host machine.

Achieved an 80% autonomous recovery rate on a rigorous 10-case failure benchmark of severely broken environments (dependency clashes, lockfile errors, missing env vars).

Proved a massive AI thesis: Giving an LLM a massive token context window is useless for debugging local dependencies. Verification is infinitely more important than generation.