Sub-Agent Orchestration Patterns FAQs

Question 1

What is the main benefit of using sub-agents in Claude Code?

Accepted Answer

The primary benefit is context hygiene. By delegating tasks to sub-agents, you prevent verbose tool outputs (like build logs) from bloating your main conversation, keeping the context window focused on high-level logic.

Question 2

Why does my sub-agent keep asking for Bash approval during file operations?

Accepted Answer

This happens when the agent has 'Bash' in its tools list, causing it to use shell commands like heredocs. Removing 'Bash' from the sub-agent's configuration and forcing it to use the 'Write' tool eliminates these redundant prompts.

Question 3

How do I make Claude Code recognize a new custom sub-agent?

Accepted Answer

Custom agents must be defined in .claude/agents/ (project-level) or ~/.claude/agents/ (user-level). Crucially, you must restart your Claude Code session for new agents to be discovered and appear in the /agents menu.

Question 4

Which model does the Explore sub-agent use?

Accepted Answer

The built-in Explore agent is optimized for speed and low-latency, typically utilizing the Claude 3.5 Haiku model for fast, read-only codebase analysis.

Question 5

Can I resume a conversation with a sub-agent?

Accepted Answer

Yes. When a sub-agent completes a task, it returns an agentId. You can use the command 'Resume agent [id]' to continue the interaction with all previous context preserved.

Sub-Agent Orchestration Patterns

Key Features

Use Cases

Sub-Agent Orchestration Patterns

Key Features

Use Cases