multi-agent-orchestrationlisted

# Multi-Agent Orchestration Advanced patterns for coordinating multiple Claude agents using Claude Code's built-in tools. ## Core Concepts ### Coordinator Pattern One Claude (the coordinator) orchestrates multiple worker Claudes: ``` Coordinator (restricted tools: Read, Agent, SendMessage, TaskStop) ├── Worker A (background, restricted tools) ├── Worker B (background, restricted tools) └── Worker C (background, restricted tools) ``` **Why `Read` is in the coordinator's set:** the coordinator's whole job is the Synthesis phase, and synthesis is a read operation — it consumes worker findings and any files those workers wrote. A coordinator without `Read` cannot perform the step that justifies its existence. Restrict the coordinator's *write* tools (no Edit/Write/Bash — delegate those to workers), not its ability to read. **Workflow phases:** 1. **Research** — Launch parallel workers to explore and gather information 2. **Synthesis** — Coordinator reads all findings, forms a unified understanding 3. **Implementation** — Launch workers with specific, synthesized instructions 4. **Verification** — Launch verification workers to test the implementation **Critical rule:** The coordinator MUST synthesize findings before delegating implementation. Never pass raw research output to an implementation worker — always digest it first. ## Tools ### Agent (Spawn Workers) ``` Agent: description: "short description" # required — 3-5 words prompt: "detailed task instruc