Agent Development

Autonomous agents execute multi-step tasks without manual intervention. Create a standards-compliant agent by defining a markdown file with YAML frontmatter (name, description, model, tools) and a system prompt, then generate the JSON manifest conforming to Claude Code deployment requirements.

Agent configurations require YAML frontmatter specifying name, description, model, color, and tools, followed by a system prompt and triggering examples. The output is a deterministic JSON structure with identifier, whenToUse, and systemPrompt fields validated against Claude.md constraints.

Yes. Design subagents with specific triggering conditions and system prompts that define when and how they operate. Each agent receives its own tool access and execution context, enabling intelligent delegation of distinct workflows.

Create agents/code-reviewer.md with frontmatter defining the agent name and description (e.g., 'Use this agent when code needs quality review'), then add system prompt logic and example triggers. Deploy using the generated JSON manifest.

Agent configurations validate field formats and length constraints according to Claude.md standards, ensuring the JSON manifest structure (identifier, whenToUse, systemPrompt) conforms to Claude Code plugin deployment requirements.

No. Agent definitions include example-based triggering that automatically initiates execution when conditions match. The deterministic JSON manifest handles deployment across Claude Code plugins without manual task orchestration.