How It Works

Instructions for integrating Cognitive Skills with your favorite AI agents

How-To Guide: Using Cognitive Skills with AI Agents

This guide explains how to take the entities you've created or found in the Cognitive Library and plug them into common AI agents and orchestrators.

1. Google Gemini (System Instructions)

Gemini works best when you provide the Markdown content of a Competency or Skill directly into its System Instructions.

  • Manual: Copy the Markdown from the "View Details" page and paste it into the "System Instructions" field in Google AI Studio or the Gemini web interface.
  • API: Send the Markdown as the system_instruction parameter in the generateContent request.
  • Tool Use: If your Skill uses Tools, define them using the Tool Schema (JSON) provided in the Studio's sidebar.

2. OpenAI (GPTs & Assistants)

OpenAI agents benefit from the structured nature of our library to reduce hallucinations.

  • Custom GPTs: Zip your project (Export ZIP) and upload it to the "Knowledge" section of your GPT, or copy the Skill Markdown into the "Instructions" box.
  • Assistant API: Upload the Skill Markdown files as code_interpreter files or inject them into the instructions field.
  • Tools: Convert our Tool definitions into OpenAI's functions format (supported natively in our Exports).

3. Anthropic Claude (System Prompts)

Claude's long context window and reasoning capabilities make it ideal for Competencies.

  • System Prompt: Inject the full text of a Competency into the system prompt.
  • Artifacts: Use the Library's generated JSON-LD as a reference "Artifact" that Claude can query during a conversation.
  • Orchestration: Direct Claude to follow the "Cognitive Markdown" workflows (e.g., > ACTION, ? DECISION) as strict procedural guardrails.

4. Cursor (The IDE)

You can use the library as a source of truth for your AI-assisted coding.

  • .cursorrules: Copy your Skill definitions into a .cursorrules file in your repository. This makes the agent follow your defined "Best Practices" for those specific tasks.
  • @docs: Add the URL of your hosted Library (or individual registry.json) to Cursor's @docs feature to make the agent aware of your available Tools and Skills.

5. Orchestrators (n8n)

For automated workflows, use the library as a configuration layer.

  • AI Agent Node: Use the "AI Agent" node and connect a "Knowledge" source that points to your Library's exports.
  • Dynamic Context: Use an HTTP Request node to fetch a Skill's Markdown content dynamically from your GitHub repository based on the task type, then feed it into the Agent's prompt.
  • Tool Mapping: Map our Tool definitions directly to n8n "Tool" nodes for seamless API execution.

Technical Maintenance

This guide is maintained as a GitHub-backed Markdown file. You can contribute to it or improve the instructions for new agents by submitting a Pull Request to the repository.

Want to improve this guide?
Edit it on [GitHub](https://github.com/modellingDH/competencies-skills/blob/main/docs/guides/AI_AGENT_INTEGRATION.md)