Safe Debug

When you paste a CUDA OOM error or a NaN loss traceback into Claude and want a careful diagnosis before any code gets touched, this is the skill to reach for. It treats debugging as a distinct phase from patching: you get a diagnosis, a minimal patch plan, and explicit approval gates before anything mutates. The design assumes you're working on research code where a quick fix might quietly break experimental comparability, so it separates "this stops the error" from "this preserves what you were measuring." If you've ever had an AI confidently refactor your training loop when you just wanted to know why the checkpoint won't load, this is the corrective.

skills.sh installs: 353K
GitHub stars: 486
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add lllllllama/rigorpilot-skills --skill safe-debug --agent claude-code

Installs into .claude/skills of the current project.

Files

safe-debug

Use this as the Rigor Debug / Rigor Audit skill. The installed slug remains safe-debug for compatibility.

Use the shared operating principles in ../ai-research-reproduction/references/agent-operating-principles.md; this skill should guide conservative diagnosis without blocking the model from finding the local root cause.

When to apply

The user provides a traceback, terminal error, or concrete training or inference failure symptom.
The user wants diagnosis, root-cause narrowing, and minimal patch suggestions before code is changed.
The user wants a safe debug flow with explicit human approval before mutation.