team-dispatch_skill

Overview

This skill orchestrates multi-agent workflows by automatically decomposing requests into DAGs, dispatching tasks to subagents, and tracking durable task state until delivery. It handles parallel execution, retries, fallbacks, and completion notifications to ensure end-to-end delivery. The skill includes installation and environment checks to prepare agent workspaces and templates.

How this skill works

It inspects the incoming request to determine complexity and selects an appropriate decomposition template (development, research, full-stack, analysis, or content). Tasks are modeled as nodes in a DAG with explicit dependencies; ready nodes are dispatched via sessions_spawn and tracked in durable JSON project files. On every subagent completion event the skill runs an immediate scheduling loop: mark task done, inject upstream results into downstream prompts, dispatch newly ready tasks, handle retries/fallbacks, and archive when all tasks finish.

When to use it

• When a request requires coordinated work across multiple agents with dependencies or parallel branches.
• When you need durable task tracking that survives context compaction or restarts.
• When you want automatic retries, fallback agents, or configurable onFailure behaviors.
• When you need event-driven, immediate cascade dispatch on task completion.
• When you require final delivery notification and project archiving.

Best practices

• Run the setup script and environment checks before first use to ensure workspaces, agent templates, and configs exist.
• Model tasks with clear dependencies and timeouts and set sensible retry limits and onFailure policies (block/skip/fallback/manual).
• Use checkpoints for XL projects to pause for user review instead of automatic dispatch.
• Dispatch ready parallel tasks simultaneously rather than serializing to avoid bottlenecks.
• Provide useful upstream results in task descriptions so downstream agents have context for accurate responses.

Example use cases

• Orchestrate a product → coder → tester → writer pipeline for a new feature delivery with automatic handoffs and final archive.
• Run a research → product → writer DAG to produce a market analysis report with visualizations and a final deliverable link.
• Execute a complex data analysis DAG where multiple analyses run in parallel and results are aggregated into a single report.
• Create a demo project to validate the multi-agent loop and confirm automatic archiving and notification behavior.