Self-hosted alternatives to safe-debug

S

safe-debug

Rigor Debug / Rigor Audit skill for deep learning research work

safe-debug is an agent skill from the rigorpilot-skills collection for diagnosing failures in deep learning research code. It is the Rigor Debug and Rigor Audit skill, and its installed name is kept for compatibility. It is used when a user pastes a traceback, a terminal error, a CUDA out-of-memory message, a checkpoint loading failure, a tensor shape mismatch or another concrete training or inference failure. The skill aims for conservative diagnosis: it narrows the root cause, explains the likely local fault and proposes minimal patch suggestions before any code is changed. It is built to keep a human in the loop, so mutation of the code or the environment needs explicit approval, and it does not block the model from finding the real cause on its own. The skill points the assistant to the shared operating principles for the research repository before it works. It is useful for a researcher who wants a fast, cautious explanation of a failure and a small, reviewable fix rather than a sweeping rewrite. It should not be used for planning new experiments or for broad refactoring. The skill passes Gen Agent Trust Hub, Socket and Snyk on skills.sh, and it has about 319,000 installs. Install it with npx skills add https://github.com/lllllllama/rigorpilot-skills --skill safe-debug.