Sub-Agent Patterns & Workflows FAQs

Question 1

How do I stop Claude from asking for Bash approval every time a sub-agent runs?

Accepted Answer

You can limit the sub-agent's toolset to only 'Write' and 'Edit' tools to force it away from shell commands, or place strict instructions at the very top of the agent's system prompt to prioritize dedicated file-writing tools over Bash heredocs.

Question 2

What is the difference between the Explore and Plan agents?

Accepted Answer

Explore is a fast, lightweight agent optimized for read-only codebase searching using the Haiku model. The Plan agent is specialized for research and information gathering specifically while Claude is in its dedicated 'plan mode'.

Question 3

Why is my new sub-agent not showing up when I type /agents?

Accepted Answer

Claude Code loads agents at the start of a session. If you create a new agent file, you must restart your Claude Code session for the agent to be discovered and become available for invocation.

Question 4

Can I use custom sub-agents across different projects?

Accepted Answer

Yes. By placing your agent configuration files in the ~/.claude/agents/ directory, they become 'User-level' agents available globally for all projects you work on using Claude Code.

Question 5

Why should I use sub-agents instead of the main Claude session?

Accepted Answer

Sub-agents provide context hygiene. They keep your main conversation clean by handling verbose tool outputs (like build logs or git status) and intermediate reasoning in isolated environments, returning only a summary to the main context.

Sub-Agent Patterns & Workflows

Sub-Agent Patterns & Workflows

主な機能

ユースケース

主な機能

ユースケース