agent-development

YAML frontmatter for Claude Code agents requires specific discovery fields like name and description to ensure reliable triggering. You must also configure the model, color, and optional tool restrictions to define the agent's operational boundaries accurately.

System prompts for autonomous agents should be written in second-person instruction formatting. Structure them with clear responsibilities, process steps, expected output format, and explicit edge-case handling to ensure consistent, high-quality behavior.

Triggering relies on embedding a “When to invoke” section with representative prose scenarios in the agent configuration. This allows the harness to accurately dispatch the agent based on the description and YAML frontmatter discovery fields.

Yes, you can restrict specific tools within the YAML frontmatter configuration. Defining optional tool restrictions helps constrain the agent's operational scope and ensures it only accesses necessary components during autonomous task execution.

Incorrect triggering often results from poorly structured YAML frontmatter or missing discovery fields. Ensure your configuration includes a robust description and a “When to invoke” section with representative prose scenarios for accurate harness dispatch.

Yes, specifying the model choice and color is part of the standard agent configuration blueprint. These elements, alongside the identifier and triggering description, ensure the agent integrates properly into the Claude Code environment.