What problem does it solve?
Deep, multi-layered problems (research → expand → verify → synthesize) can exhaust a single agent's context window. This Skill delegates work through a tree of sub-agents where each level gets a fresh context window, so intermediate state never pollutes the top-level agent.
Core Features & Use Cases
- Nested delegation up to depth 5: A
nested-coordinator agent decomposes the problem with TodoWrite, then spawns children that can themselves spawn children via the Task tool.
- Context isolation per level: Children return ~200-token structured summaries instead of full transcripts, keeping the parent's context clean.
- Depth guard rails: A
pre-task hook enforces a default cap of 4 levels (configurable in claude-flow.config.json) and returns a typed NESTING_DEPTH_EXCEEDED error at the limit.
- Tree-shaped cost attribution: A
post-task hook persists parent_agent_id and depth to AgentDB for per-tree spend analysis.
- Use Case: A recursive investigation where a coordinator spawns research agents, each of which spawns verification and synthesis agents, with only summaries climbing back to the root.
Quick Start
Ask the agent to decompose your problem statement into a nested sub-agent tree using the nested-coordinator, then review the proposed spawn tree before approving deep work.
