Efficiently Scale Your Agent Task Execution

The subagent-dispatch skill solves the "coordination overhead" problem often found in multi-agent workflows. Instead of manually deciding how to split a complex task or letting an LLM over-spawn unnecessary agents, this skill analyzes your task and determines the mathematically optimal execution strategy: SOLO, DUAL, or SWARM-N.

What it does

This skill acts as an intelligent router for your AI agent. It evaluates work based on dependency chains and output volume to provide a ready-to-run execution plan. It supports:

• SOLO: Sequential execution for tasks with high dependency.
• DUAL: Parallel execution for independent halves (e.g., Backend vs. Frontend).
• SWARM-N: Fan-out execution for repetitive tasks across multiple files or repositories (up to 7 agents).

Why use this skill?

Simply asking an AI to "do this in parallel" often results in context leakage or poorly defined sub-tasks. This skill generates self-contained, context-free prompts for every subagent, ensuring they have exactly what they need to succeed without parent-context "noise." It also enforces strict token budgets, preventing runaway costs while minimizing wall-clock time.

Output format

You receive a structured dispatch order including the reason for the split and a PASTE THIS block. You can copy-paste the resulting instruction directly into your Task tool or terminal to trigger the parallel work immediately.