subagent-orchestrationlisted

## Routing Guidance - **Hub Skill:** Combines agent definition authoring (DESIGN) and subagent execution (ORCHESTRATE). - **Exclusion:** Do NOT use for multi-agent architecture decision-making (supervisor vs swarm) — read `references/patterns-reference.md` for pattern selection. - Do NOT use for single-agent tasks, simple scripts, or non-agent workflows. # Subagents ## Decision Router Pick the execution mode by task scale and shape — not by which mechanic is most familiar. The runtime exposes more than one primitive now; the orchestrator's first job is selecting the right one. ### Execution-mode selection | Task scale | Right primitive | When to choose it | |---|---|---| | Trivial — 1 file, <10 lines, or a single search/read | Inline in the main context | Quick edit, single lookup, glance-check | | Non-trivial single-context — 3–10 files, single methodology, side task | Subagents | Result returns and is synthesized in this conversation; orchestration shape is one-off | | Multi-stage with fan-out → verify → synthesize | Orchestration script | Repeatable shape worth codifying; structured-output schemas; adversarial verification across N independent agents | | Codebase-wide or many-file or multi-methodology | Orchestration script with explicit phase structure | Coordination outgrows a handful of subagent spawns per turn | | Long-running with external triggers | Orchestration script + recurring checks + push channels | Reacts to CI, alerts, scheduled events; survives idle t