Audit the local Claude Code setup end to end and report before changing anything. Plugins installed vs enabled, MCP servers configured vs connected vs unauthenticated, skills installed vs actually invoked (counted from the transcripts), duplicate tool servers, hooks and whether any mirrored hook copies are byte-identical, brain index vs files, settings model vs the model actually running.
安装
npx skills add Sagargupta16/agent-brain --skill harness-auditnpx skills add Sagargupta16/agent-brain --skill harness-audit -a claude-code -gnpx skills add Sagargupta16/agent-brain --skill harness-audit -a codex -ggemini skills install https://github.com/Sagargupta16/agent-brain.git --path skills/harness-auditgh skill install Sagargupta16/agent-brain harness-auditgit clone --depth 1 https://github.com/Sagargupta16/agent-brain /tmp/Sagargupta16__agent-brain
mkdir -p ~/.claude/skills ~/.agents/skills
cp -R /tmp/Sagargupta16__agent-brain/skills/harness-audit ~/.claude/skills/harness-audit # Claude Code
cp -R /tmp/Sagargupta16__agent-brain/skills/harness-audit ~/.agents/skills/harness-audit # Codex, Cursor, Gemini CLI, Copilot, OpenCode