Use when designing a goal-oriented agent loop or reviewing one for runaway risk - build gate, machine-decidable goal, anti-Goodhart boundaries, five failure modes. Triggers on "design a loop", "review this loop", "will this loop run away".
Use when running OpenAI Codex via the codex CLI (copilot CLI fallback) for code analysis, refactoring, or editing. Triggers on "run codex", "codex analyze", "ask codex", "copilot with codex".
Use when running Google Gemini via the native Gemini CLI for code analysis, refactoring, or editing. Triggers on "run gemini", "gemini analyze", "ask gemini".
Use when running a local Qwen model (via llama.cpp and the pi agent) for code analysis, refactoring, or editing - local inference, no API cost. Triggers on "run qwen", "qwen analyze", "ask qwen", "local model task".
Use when running Anthropic Claude Sonnet via the native claude CLI (headless) for code analysis, refactoring, or editing. Triggers on "run sonnet", "sonnet analyze", "ask sonnet".
Use when converting a plan, design doc, brainstorming output, or discovery doc into a structured PRD saved to the backlog. Triggers on "create PRD", "create PRD from plan", "make PRD", "convert to PRD", "save plan as PRD".
Use when breaking a PRD into granular, sequenced implementation tasks ready for the work skill. Triggers on "plan tasks", "create tasks from PRD", "implement PRD", "break down the spec".
Use when executing already-planned tasks one at a time, dispatching the implementor and committing after each. Triggers on "work on tasks", "implement tasks", "start working", "execute the plan", "do the work".