Task Delegation

The delegate_task tool lets the agent spawn subagents to handle complex or parallel work. Each subagent runs in isolation with its own context.

When to use delegation

Delegation is useful for:

• Parallel research — “Research A and B simultaneously and compare them”
• Context-heavy tasks — Tasks that would produce too much output for the main context
• Isolated debugging — Debug sessions that explore many files
• Code review — Analyzing large diffs without flooding the conversation

How it works

When the agent delegates a task:

1. A new subagent is created with a fresh conversation
2. The subagent gets a focused goal and any relevant context
3. It works independently with a restricted toolset
4. Only the final summary is returned to the main conversation

The parent conversation never sees the intermediate tool calls or reasoning — just the result.

Restrictions

Subagents have some limitations to prevent runaway behavior:

• No recursive delegation — Subagents cannot spawn their own subagents
• No user interaction — Subagents cannot ask clarifying questions
• No memory writes — Subagents cannot modify MEMORY.md or USER.md
• No messaging — Subagents cannot send messages to other platforms

Parallel tasks

You can request parallel work:

Research these three topics in parallel:
1. Current state of Rust web frameworks
2. Best practices for GraphQL authentication
3. Performance comparison of PostgreSQL vs MySQL

The agent will spawn up to 3 subagents concurrently and combine the results.

Example

You: Debug why my tests are failing in the auth module. Look at all the test files and the implementation.

Agent: I'll delegate this to a subagent to analyze all the files without flooding our conversation...
[delegates task]
[receives summary]

The issue is in `auth.test.js` line 47 — the mock isn't being reset between tests...