Skip to content

GitHub AI Integrations

This guide documents the different approaches for integrating AI agents with GitHub workflows in ontology repositories. Each approach has different trade-offs in terms of setup complexity, billing, and capabilities.

Overview

Approach Billing Model Selection Setup Complexity Best For
Dragon-AI Agent API key (project-based) Configurable Medium Custom workflows, team control
GitHub Copilot GitHub subscription GitHub-controlled Low Quick setup, GitHub-native
Claude Code Action API key or Max subscription Anthropic models Low-Medium Claude-specific features

Dragon-AI Agent

The Dragon-AI Agent approach uses custom GitHub Actions to deploy headless AI coding assistants (Claude Code or Goose) in response to issue/PR comments.

How It Works

  1. A controller invokes the agent with @dragon-ai-agent please in an issue or PR comment
  2. A GitHub Action triggers, running the AI in a containerized environment
  3. The AI reads the issue context, makes changes, and creates/updates PRs
  4. Controllers are authorized via .github/ai-controllers.json

Setup

See Set up GitHub Actions for detailed setup instructions.

Key configuration files: - .github/workflows/ - GitHub Action workflow definitions - .github/ai-controllers.json - Authorized users list - CLAUDE.md - AI system instructions

When to Use

  • Team control: You want fine-grained control over who can invoke the AI
  • Custom tooling: Your workflow requires specific MCP servers or tools
  • Project billing: You want to charge AI usage to a specific project/grant via API proxy
  • Multi-model support: You need to switch between different AI providers

Limitations

  • Requires maintenance of GitHub Action workflows
  • Setup is more involved than native integrations
  • Debugging requires checking GitHub Actions logs

GitHub Copilot

GitHub Copilot's coding agent can be assigned to issues and PRs directly through the GitHub interface.

How It Works

  1. Assign an issue to Copilot - it creates a PR to address the issue
  2. Assign a PR to Copilot - it reviews and suggests changes
  3. Copilot works within GitHub's infrastructure

Setup

  1. Enable GitHub Copilot for your organization/repository
  2. Copilot appears as an assignable user on issues and PRs

For educational users, see the GitHub Education benefits section for free Copilot Pro access.

When to Use

  • Quick setup: You want to start using AI agents immediately
  • GitHub-native: You prefer staying within GitHub's ecosystem
  • Individual use: For personal repositories or small teams
  • PR reviews: Copilot excels at code review tasks

Limitations

  • Less control over model selection and behavior
  • Billing tied to GitHub subscription
  • May not support ontology-specific tooling (ROBOT, OWL tools)
  • Configuration options are limited compared to custom approaches

Ontology-Specific Considerations

For ontology repositories, Copilot may need additional guidance:

  • Include clear instructions in repository documentation
  • Copilot may attempt to run tools locally before using ODK wrappers
  • Add prominent warnings in README/CLAUDE.md about using ODK containers

Claude Code Action

Anthropic's official Claude Code Action provides a streamlined way to run Claude Code in GitHub Actions.

How It Works

  1. Trigger via issue/PR comments (configurable trigger phrase)
  2. Claude Code runs with access to repository contents
  3. Can create commits, PRs, and respond to comments

Setup

Install via Claude Code: 

claude /install-github-app

Or manually add the GitHub Action to your repository.

Billing options: - API key (pay-per-use via Anthropic API) - Claude Max subscription (included usage)

When to Use

  • Claude-specific features: You want access to latest Claude capabilities
  • Simple setup: Official action with maintained support
  • Flexible billing: Choose between API or subscription billing
  • Anthropic ecosystem: Already using Claude for other workflows

Limitations

  • Limited to Anthropic models
  • Less customization than Dragon-AI approach
  • Requires Anthropic API key or Max subscription

Comparison for Ontology Repositories

For ontology curation workflows, consider these factors:

Tool Access

Tool Dragon-AI Copilot Claude Code Action
ROBOT via ODK Yes (configurable) Limited Yes (configurable)
OWL-MCP Yes No Yes
Custom MCP servers Yes No Yes
Web search Yes Limited Yes

Starting out / Experimentation: - Use GitHub Copilot for quick wins on simple issues - Low barrier to entry, good for learning

Production ontology curation: - Use Dragon-AI Agent or Claude Code Action - Better tool integration and customization - Project-based billing for grant compliance

Mixed team (technical + non-technical): - Dragon-AI Agent with clear controller authorization - Provides guardrails while enabling AI assistance

Configuration Files

Regardless of which approach you use, these files help guide AI behavior:

File Purpose
CLAUDE.md System instructions for Claude-based agents
.goosehints Instructions for Goose (often symlinked to CLAUDE.md)
.github/copilot-instructions.md Instructions for GitHub Copilot
.github/ai-controllers.json Authorized users for Dragon-AI