agent-development

To create a Claude Code agent, you define a markdown file with valid YAML frontmatter including name, description, model, color, and optional tools fields. This Skill provides structured guidance to ensure your agent configuration is packaged correctly and behaves consistently across development workflows.

The best way to write triggers for Claude Code plugins is to write strong descriptions with example-driven activation criteria. This ensures the agent learns exactly when to activate based on specific user contexts and inputs, preventing unintended triggering during development workflows.

You should structure a system prompt using second-person instructions that outline process steps, quality standards, output format, and edge-case handling. This approach shapes predictable behavior for autonomous assistants within your plugin architecture.

Yes, this Skill includes validation for agent configuration files. You use the provided validator script and guidance to confirm your agent files are well-formed with valid frontmatter and trigger as intended before deploying them.

Yes, you can restrict tool access when building a Claude Code agent by defining optional tools fields within the YAML frontmatter. This allows you to limit the specific actions and permissions available to the agent during execution.