creating-claude-agents

Create a Claude agent by defining required frontmatter fields (name, description, model, allowed-tools, agentType), adding a mandatory H1 heading, defining a persona, and structuring content sections (Instructions, Process, Examples, Guidelines). Validate your agent against the Claude agent schema to ensure conformance before deployment.

A Claude agent is a persona-driven interface with restricted tool access and custom frontmatter; a Skill is a reusable knowledge component; a slash command is a shorthand trigger. Agents require full schema validation and persona definition, while Skills and slash commands serve different architectural roles within Claude Code.

Define the allowed-tools field in your agent frontmatter to specify exactly which tools the agent can access. This constrains functionality to your intended use case and improves security and performance by preventing unnecessary tool exposure.

Model selection depends on your agent's complexity and latency requirements. The metadata field in frontmatter accepts model parameters; choose based on your task scope, tool dependencies, and performance constraints documented in agent creation best practices.

Agent schema validation errors typically stem from missing or malformed frontmatter fields, missing H1 heading, incomplete persona definition, or missing required content sections (Instructions, Process, Examples, Guidelines). Review each field against the Claude agent schema specification and ensure all mandatory elements are present.

Yes. Apply agent creation best practices to troubleshoot validation errors, improve persona clarity, refine tool access restrictions, optimize model selection, and restructure content sections. Validation against the schema ensures your refined agent meets technical and functional requirements.