Skip to main content

🧠 Customizing Skills

Beak AI Studio allows you to extend agent capabilities by adding custom skills. Skills can be written in Python, exposed via REST APIs, defined as LLM prompts, or integrated using the Model Context Protocol (MCP).

Note: Custom skills are now created using a streamlined two-step workflow. First, define the skill (type, name, description, usage instructions). Then, configure technical details (code, OpenAPI spec, prompt, authentication) during agent integration. See Custom Skills for a full overview.

Types of Skills​

  • Pre-built Skills: Provided by Beak AI for common tasks (search, summarization, etc.).
  • Custom Skills: Create you own skill using pre-defined types and register it in Studio.
  • MCP Skills: Use the Model Context Protocol for distributed, language-agnostic skills.

Creating a Custom Skill​

  1. Create the skill in Studio:

    • Choose "Generate and Execute Python Code" as the type, provide a name, description, and usage instructions. (The instructions determine what code needs to be generated)

  2. Test the skill in the Studio playground.

Integrating REST API Skills​

  1. Create the skill in Studio:
    • Choose "REST API" as the type, provide a name, description, and usage instructions.
  2. Configure during agent integration:
    • Upload or paste the OpenAPI specification for the API.
    • Provide authentication details (API keys, OAuth, etc.) as needed.
  3. Test the skill in the Studio playground.

Creating an LLM Prompt Skill​

  1. Create the skill in Studio:
    • Choose "LLM Prompt" as the type, provide a name, description, and usage instructions.
  2. Configure during agent integration:
    • Enter the prompt template and select the response parsing format (JSON or string).
  3. **Test the skill in the Studio playground.

Example prompt to provide during agent integration:

Summarize the following customer feedback in one sentence: {text}

Using MCP Servers​

  • Deploy your skill as an MCP server (see MCP documentation).
  • Register the MCP server in Studio and map its tools as skills.
  • Benefit from language-agnostic, distributed skill execution.

Best Practices​

  • Use clear, versioned schemas for all skill inputs/outputs.
  • Add docstrings and comments for maintainability.
  • Test skills thoroughly in the Studio playground before production use.
  • Document dependencies and integration steps.

For more advanced customization, see the Studio Skills Documentation and MCP Skills Guide.