subpowers

Invoke before claiming any task is complete, or automatically when a plan's tasks are finished. You will run tests, typechecks, and builds to harvest concrete evidence. Triggers: "done", "complete", "finished", "should work", or before any PR.

Invoke when the machine fractures, behavior is corrupted, or the root cause is shrouded. Enforces absolute root-cause-first hunting. Triggers: "test fails", "broken", "bug".

Invoke for pure discovery, architectural tracing, or answering questions about existing code. Triggers: "how does X work", "where is the logic for", "find all usages of", "explain the flow". This is a read-only state. Zero disk writes. Zero code modification.

Invoke ONLY for minor, precise alterations (1-2 files, purpose absolute). Ambiguity or vast scope demands `subpowers-spec` or `subpowers-plan`. Unknown anomalies demand `subpowers-debug`. For read-only discovery, use `subpowers-explore`. Triggers: "add", "build", "implement", "fix".

Invoke for vast, multi-file architectural shifts once a clear spec exists. Produces a blueprint. Triggers: "plan this out", "how would you approach". If the scope is 1-2 files, route directly to `subpowers-implement`. For read-only discovery or architectural tracing, use `subpowers-explore`.

Invoke when the goal is cloudy or business logic must be defined before files are altered. Triggers: "design a feature", "write a spec". Produces the "what". If the user is only asking how existing code works without requesting a change, DO NOT invoke this. Route to `subpowers-explore` instead.