Agent Development FAQs

Question 1

What models can these agents use?

Accepted Answer

Agents can use Claude Sonnet, Opus, Haiku, or 'inherit' the model from the parent process, which is the recommended default for most use cases.

Question 2

How do I trigger an agent in Claude Code?

Accepted Answer

Agents are triggered based on the 'description' field in the YAML frontmatter. This field should include specific conditions and multiple examples showing context, user requests, and expected assistant responses.

Question 3

What is the recommended file structure for an agent?

Accepted Answer

Agents should be saved as .md files in the 'agents/' directory of your plugin. They require YAML frontmatter (name, description, model, color) followed by a structured system prompt in the markdown body.

Question 4

What is a Claude Code Agent?

Accepted Answer

Agents are autonomous subprocesses that handle complex, multi-step tasks independently. Unlike simple commands, agents are designed for autonomous work and can be triggered automatically by specific conditions.

Question 5

Can I limit the tools an agent uses?

Accepted Answer

Yes, the 'tools' field in the frontmatter allows you to restrict an agent to a specific set of capabilities, such as read-only access or specific shell commands, following the principle of least privilege.

Agent Development

主な機能

ユースケース

Agent Development

主な機能

ユースケース