create-subagent

To generate a single-purpose Claude Code sub-agent, you provide a description or requirements document. The Skill runs a multi-stage pipeline including an adaptive interview, classification, and generation via Sonnet to scaffold the agent definition. It then deploys the validated file to the .claude/agents/ directory.

AI agent scaffolding is the automated creation of single-purpose agent definitions with diagnostics and permissions setup. You need it when you want dedicated Claude Code agents for specific tasks like security vulnerability analysis, ensuring they are properly configured and validated before deployment.

Generated AI agents are validated using the integrated anthropic-validator. This automated validation step ensures the scaffolded agent definitions meet code quality standards and adhere to conventions before being deployed to the .claude/agents/ directory.

Yes, the scaffolding process references required templates like template-single-agent.md and agent-conventions.md. These templates structure the agent output, ensuring the generated definitions include proper diagnostics, permissions, and content guidance.

The best way to define permissions is through an automated scaffolding pipeline that classifies requirements and sets up tool permissions. This ensures the generated single-purpose agents have validated configurations before deployment.

When scaffolding agents for security vulnerability analysis, the process requires specific reference files like decision-framework.md and content-guidance.md to be present. Without these structured inputs, the generated agent output may lack proper validation and permissions setup.