A disciplined loop for hard bugs, flaky tests and performance regressions - build a red-capable feedback loop first, reproduce and minimize, rank falsifiable hypotheses, instrument one variable at a time, then repair and remove the probes.
Install
npx skills add ehnruiz/claude-skills --skill diagnosing-bugsnpx skills add ehnruiz/claude-skills --skill diagnosing-bugs -a claude-code -gnpx skills add ehnruiz/claude-skills --skill diagnosing-bugs -a codex -ggemini skills install https://github.com/ehnruiz/claude-skills.git --path skills/diagnosing-bugsgh skill install ehnruiz/claude-skills diagnosing-bugsgit clone --depth 1 https://github.com/ehnruiz/claude-skills /tmp/ehnruiz__claude-skills
mkdir -p ~/.claude/skills ~/.agents/skills
cp -R /tmp/ehnruiz__claude-skills/skills/diagnosing-bugs ~/.claude/skills/diagnosing-bugs # Claude Code
cp -R /tmp/ehnruiz__claude-skills/skills/diagnosing-bugs ~/.agents/skills/diagnosing-bugs # Codex, Cursor, Gemini CLI, Copilot, OpenCode