AI_AGENT_SETUP.md

AI Agent Setup Instructions

This repository has been configured with a GitHub Actions workflow for AI agent integration using the Dragon AI Agent.

Files Created

1. .github/workflows/ai-agent.yml - GitHub Actions workflow that triggers the AI agent on issue/PR mentions
2. .config/goose/config.yaml - Configuration for the goose AI agent
3. .goosehints - Symlink to CLAUDE.md containing project-specific instructions

Required Repository Secrets

To complete the setup, you need to configure the following secrets in your GitHub repository settings:

Navigate to: https://github.com/YOUR_USERNAME/kg-alzheimers/settings/secrets/actions

Required Secrets:

1. ANTHROPIC_API_KEY

   • Your Anthropic API key for Claude access
   • Get from: https://console.anthropic.com/

2. CBORG_API_KEY

   • API key for the CBORG LiteLLM proxy
   • This enables access to multiple AI models through a unified interface

3. PAT_FOR_PR

   • Personal Access Token for GitHub operations
   • Needs permissions: contents:write, pull-requests:write, issues:write
   • Generate at: https://github.com/settings/tokens

How It Works

• When someone mentions @alzassistant in issues, PRs, or comments, the workflow triggers
• The AI agent can read the context, understand requests, and create pull requests with changes
• The agent uses the instructions in .goosehints (linked to CLAUDE.md) to understand the project
• Fallback controller is set to jtr4v for permissions management

Testing

Once secrets are configured, test by:

1. Creating an issue
2. Commenting with @alzassistant help me with X
3. The agent should respond with a PR addressing the request

Configuration Notes

• Uses anthropic/claude-sonnet model via CBORG proxy
• Agent name: "alzassistant"
• Branch prefix: "alzassistant"
• Robot version: v1.9.7