← ClaudeAtlas

dev-multi-agentlisted

Breaks a large, complex task into independent pieces and works on multiple parts simultaneously, then brings the results together. Use for big features or refactors that touch many files — it dramatically reduces the time needed for large tasks.
talentedgeai/infinite-leverage · ★ 0 · AI & Automation · score 60
Install: claude install-skill talentedgeai/infinite-leverage
# Developer: Multi-Agent Delegation Use for complex tasks that can be decomposed into independent work items. ## Wave Execution Model ``` Wave 1: [Agent A] [Agent B] [Agent C] ← independent, parallel ↓ ↓ ↓ done done done ↓ Wave 2: [Agent D] [Agent E] ← depends on Wave 1 outputs ``` Waves are sequential — Wave 2 starts only after Wave 1 is fully resolved. Within a wave, all agents run concurrently and independently. ## Sub-Agent Prompt Template Every sub-agent prompt must include: ``` You are working on [specific scope]. Your goal: [single deliverable]. Context: - [relevant background] - [error messages / failing tests / requirements] Constraints: - Do NOT modify [files/systems outside scope] - Do NOT ask clarifying questions — make the best call and document your reasoning When done, return: - Decision made / root cause - Files created or modified (path + one-line description) - Any caveats or follow-up items ``` ## When to Use Parallel Dispatch | Situation | Action | |-----------|--------| | 2+ independent failures/tasks | Dispatch one agent per domain | | Tasks share state or depend on each other | Sequential agents or single agent | | Single problem, unclear root cause | Single agent investigates first | | Agents would edit the same files | Do NOT dispatch in parallel | ## Coordinator Responsibilities - Decompose task into independent domains before Wave 1 - Craft complete, s