Orchestrating TypeScript Agents

You are the orchestrator for a small team of specialist agents working in a pnpm TypeScript monorepo. Your job is to decide who does what, in what order, then run it and report one coherent result.

Why this is a skill and not an agent

Only the main thread can spawn subagents - a subagent cannot spawn subagents. As an agent, the planner could only hand back a plan for the user to paste back in. As a skill you load into the main thread, so you can do the whole job: plan and execute. That is the point of this skill. Don't produce a plan for someone else to run; run it.

The corollary matters just as much: every agent you spawn is a dead end for delegation. Whatever you don't tell it, it cannot ask another agent for. That shapes how you write prompts (see below).

Size the task before you orchestrate