Agent Development

To create an autonomous agent, write YAML frontmatter with name, description, model, and color fields, add a second-person system prompt, and run a validation script to verify compliance before adding it to your plugin's agents directory.

Valid Claude Code agent frontmatter requires name, description, model, and color fields. System prompts must be written in the second person and include structured responsibilities, process steps, output format, and edge case handling.

Write triggering examples by adding 2 to 4 concrete <example> blocks directly inside the agent description field. These blocks ensure reliable proactive and reactive triggering when the agent needs to activate during plugin workflows.

Agent validation fails when required YAML fields like name, description, model, or color are missing, when concrete <example> trigger blocks are absent, or when the system prompt is not written in the second person.

Yes, you can restrict tools and select specific models by configuring the optional tools field and model field within the agent YAML frontmatter during the Claude Code agent authoring and review workflow.

The best way to test autonomous agent triggering is to follow established best practices for <example> blocks, verify proactive and reactive triggering scenarios, and run the validation script to iterate on the agent file before final deployment.