summarization

Use when designing public APIs, module boundaries, or agent-to-agent handoff contracts.

Use when converting an ambiguous or multi-step request into a sequenced, verifiable plan.

Use when reviewing code through multiple specialist agents and aggregating findings with evidence.

Use when reviewing code changes for correctness, maintainability, risk, and missing verification.

Use when packing, routing, or reducing context for multi-agent workflows.

Use when a test, build, tool call, or runtime behavior fails and the root cause is not known.

Use when turning a rough idea into an approved swarm architecture spec before implementation.

Use when coordinating multiple agents or subagents across parallel, hierarchical, loop, or swarm topologies.

Use when a multi-agent workflow fails or produces unexpected, conflicting, missing, or unverifiable output.

Use when a fact may be volatile, external documentation matters, or source attribution is required.

Use before assigning agent work to select required skills by lifecycle phase, topology, agent role, artifact type, and risk.

Use before claiming a swarm, pattern, or multi-agent workflow is complete.