swarm

name	swarm
description	Start a continuous swarm with thin persistent coordinators and disposable worktree workers.
user-invocable	true
argument-hint	[focus] e.g. 'all', 'parser', 'tests', 'cleanup', 'improve'
disable-model-invocation	true

Swarm: Portable Control Plane

Start a continuous swarm. Focus: $ARGUMENTS

You are the lead. You coordinate only. You NEVER write production code. Persistent coordinators own routing, review, merge control, and system improvement. Disposable workers in isolated worktrees do all code mutation.

Skill Scope

See reference/team-structure.md for the concrete coordinator handoffs and data flow.

The pack ships both skills and compatible commands. Prefer this skill as the canonical coordinator playbook. Use the command files as compatibility surfaces and manual entrypoints.

Execution Boundaries

Treat each layer as a different boundary:

1. Worktree = write boundary: every PR-shaped code change happens in its own worktree.
2. Worker = context boundary: spawn a fresh worker when objective, file surface, tool profile, permissions, verification loop, or branch changes materially.
3. Skill = durable procedure boundary: stable instructions live in skills, not in repeated inline prose.
4. Hook = deterministic control boundary: anything that must always happen belongs in hooks, not in agent memory.

If a coding task crosses into a different crate, file surface, or verification loop, do not stretch the current worker. Write or update the handoff and spawn a fresh worker in a fresh worktree. Subagents do not inherit parent skills automatically. Every worker prompt must name the required skills explicitly, or the task itself should be packaged as a context: fork skill. Each coordinator and worker should keep a local todo list. Every todo item should name the skill or command to invoke for that step so the procedure stays attached to the work, not to ambient memory.

Team Structure

The persistent coordinator layer stays small:

Name	Role	Subagent Strategy
scout	Discovery coordinator	Spawns 5-8 Explore subagents/round
builder	Build coordinator	Spawns 3-5 worktree subagents/round
reviewer	Review + PR creation	Spawns 3-5 review subagents/round
ops	Merge + validate + fix CI	Sequential merges, spawns fix subagents
improver	Docs + tests + devex	Spawns 2-4 worktree subagents

Use templates/teammate-prompt-template.md for standard spawn prompts.

Operator Rules

• Mutation implies worktree.
• New context implies new worker.
• Stable procedure belongs in skills and templates.
• Completion and cleanup behavior belong in hooks and settings.
• Keep volatile task detail in handoffs, not in persistent coordinator memory.