Use when the user wants to do a QA session or report multiple bugs — interactive session where bugs are reported conversationally and agents fix them in parallel
Use when executing implementation plans with independent tasks — orchestration pattern for worktree isolation, TDD discipline, and two-stage review. Referenced by execute-plan, fixit, and bugbash.
Use when the user reports a bug or issue that can be fixed without blocking their current work — backgrounds an agent in a worktree to fix and merge back without breaking stride
Use after implementing changes in an OpenSpec project to review implementation against the active change's deltas — auto-fixes confident issues, parks questions for the user
Install the anutron (claude-skills) kit into the current project — symlinks or copies skills, registers hooks, compiles CLAUDE.md from snippets.
Uninstall the anutron (claude-skills) kit from the current project — reverses everything /anutron-install did.
Reference for bash patterns that trigger Claude Code permission guardrails (static-analysis flags that allowlist rules cannot silence). Load when about to run bash with `cd <repo> && git ...`, `$(...)`, backticks, heredocs, inline `python3 -c` / `node -e` / `ruby -e` / `perl -e`, multi-line shell logic with variables/conditionals, or before allowlisting a new script path. Also covers always-prompt verbs and the `git -C` known gap.
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes — multi-agent competing hypotheses debugging