| name | subagent-dispatch-design |
| description | 需要具体的多层剧本子代理调度计划(而非仅角色列表)时使用。 |
Subagent Dispatch Design
Use this skill when the user asks how screenplay subagents should be orchestrated across layers, lanes, reviews, and human gates.
Workflow
- Confirm the primary route or target artifact.
- Confirm or infer the team mode and cast scope.
- Select a dispatch topology and phase pattern.
- Define packet flow, merge ownership, review loops, human gates, and context budgets.
- Return a plan that can be executed without hidden full-context sharing.
Output Contract
subagent_dispatch_plan: control-plane order, topology, phase ladder, lane budgets, packet schema, review loops, human gates, fallback topology, and collapse triggers.
- Prefer two-stage review where under-building and over-building are both costly.
- Keep persona lanes advisory unless the plan explicitly says otherwise.
Posture-Adaptive Guidance
When certainty = lost:
Ask: "Which part of the work cannot proceed until someone else finishes — and who is that someone?" One dependency bottleneck is the seed of a dispatch plan. Do not produce a full topology yet.
When certainty = certain:
Apply the full workflow: confirm route and cast → select topology → define packet flow and merge ownership → define review loops, human gates, and context budgets → return plan with collapse triggers.
When certainty = exploring:
Offer two topology options: a tighter sequential chain (lower parallelism, lower merge risk) and a wider parallel ring (higher throughput, higher convergence complexity). State the deciding factor.
When source = construct:
Primary activation context for this skill. Dispatch design is fundamentally a construct-mode activity: every topology decision is a structural choice with explicit cost and benefit.
When focus = event:
For event-layer screenplay work, the dispatch plan should reflect the narrative sequence: earlier structural decisions constrain later scene decisions. Topology should respect this dependency order.
References
wp.subagent-dispatch-plan
rb.subagent-dispatch-plan
ka.convergence-owner-discipline
ka.cross-protocol-referral-edges
ka.handoff-packet-discipline
ka.parallel-lane-governance
ka.subagent-context-budgeting
ka.team-topology-selection
ka.two-stage-review-loop