This skill transforms bug fixing from guesswork into a disciplined engineering process. It provides a structured framework to handle complex regressions and hard-to-reproduce bugs by enforcing a strict sequence of phases: reproduction, hypothesis generation, instrumentation, and verification.
Use this skill when the user reports a bug, describes a failure/crash, or notes a performance regression that isn't immediately obvious. It is specifically designed for 'hard' bugs that require more than a simple one-line fix.
Designed for high-capability coding agents (e.g., Claude Code, Codex, Gemini CLI) that can execute shell commands and write tests.
A well-structured debugging discipline skill that guides agents through a rigorous 6-phase diagnosis loop: build feedback loop, reproduce, hypothesise, instrument, fix, and cleanup. The bundled HITL loop template script is a human-interactive bash script that prompts for user input, so it cannot succeed in automated testing. The SKILL.md itself is thorough, well-written, and provides specific actionable instructions with ranked strategies.
Excellent skill — one of the better-structured ones seen. The 6-phase approach with explicit checkpoints (don't proceed without a loop, show hypotheses to user) is disciplined. No security concerns whatsoever. The template script is intentionally interactive and not meant for headless execution.