4 tagged decision gate, measured the same way as everything else here.
Agent
Implements one vertical slice of an AIDD build under the decision contract. Reads decisions.md, respects APPLY/INFER/INNOVATE/DEFERRED stances, verifies by running the code. Use for parallel slices during the execution phase.
Agent
Researches one decision segment exhaustively (docs, GitHub repos and code, registries, critiques, prior art) and returns a validated JSON fragment of options with pros, cons, and real sources. Use one scout per segment, launched in parallel, during the AIDD research phase.
Agent
Surveys an existing codebase before an AIDD run — what stack is in place, what conventions hold, what constraints bind, and which decisions are already closed. Use at the start of any AIDD run in brownfield mode, before research derives segments.