multi-agent-master.md

id	multi-agent-master
name	Multi-agent master (fan-out → synthesis)
when_to_use	A task with several independent facets that benefit from parallel specialist lenses converging into one synthesized result.
agents	zc-general zc-gate
skills	multi-agent-master-workflow queen-bee-schwarm-dispatch yuno-team-orchestrator yuno-team-routing
phases	phase owner_agent skills exit_criteria failure_modes Decompose zc-general yuno-team-routing multi-agent-master-workflow Task split into orthogonal sub-tasks, each with an owner lens and acceptance criteria. Overlapping sub-tasks that duplicate work or race on the same files. phase owner_agent skills exit_criteria failure_modes Fan-out (parallel lenses) zc-general queen-bee-schwarm-dispatch yuno-team-orchestrator Independent agents run concurrently, each blind to the others, returning structured findings. Shared mutable state across parallel workers; context bleed. phase owner_agent skills exit_criteria failure_modes Synthesize zc-general multi-agent-master-workflow Findings deduped and merged into one coherent result. Concatenating outputs instead of reconciling them. phase owner_agent skills exit_criteria failure_modes Gate zc-gate Independent PASS / RETRY / BLOCK on the synthesized result. Accepting the synthesis without an adversarial check.

Multi-agent master (fan-out → synthesis)

Decompose → Fan-out → Synthesize → Gate. A master controller splits work into orthogonal lenses, runs them in parallel (each independent), then reconciles into one result behind an independent gate. Uses multi-agent-master-workflow and queen-bee-schwarm-dispatch (installed), with yuno-team-* for routing/orchestration decisions.

Route in: "orchestrate / parallel audit / multiple perspectives / decompose this." Use only when sub-tasks are genuinely independent — otherwise a single-agent workflow is cheaper and safer.