Agent Development for Claude Code FAQs

Question 1

Should I specify a model for my custom agent?

Accepted Answer

It is generally recommended to use 'model: inherit'. This allows the agent to use the same model as the parent process, though you can specify 'sonnet', 'opus', or 'haiku' if the task requires specific performance or cost profiles.

Question 2

What is the difference between a command and an agent in Claude Code?

Accepted Answer

Commands are typically used for user-initiated actions with immediate output, while agents are autonomous subprocesses designed to handle complex, multi-step tasks independently.

Question 3

How does Claude know when to trigger a specific agent?

Accepted Answer

Claude uses the 'description' field in the agent's markdown frontmatter. This field must include specific triggering conditions and multiple examples with commentary to define the correct context for activation.

Question 4

What are the naming conventions for agent identifiers?

Accepted Answer

Agent names must be between 3-50 characters, use only lowercase letters, numbers, and hyphens, and must start and end with an alphanumeric character.

Question 5

Can I restrict the tools an agent has access to?

Accepted Answer

Yes, you can use the 'tools' field in the frontmatter to define an array of specific tools (e.g., ['Read', 'Grep']). If omitted, the agent has access to all available tools by default.

Agent Development for Claude Code

主な機能

ユースケース

Agent Development for Claude Code

主な機能

ユースケース