Change the actions template for claude and dragon-ai to use ODK docker image

[cmungall]

We have a discussion already for copilot

• Can we set up the copilot action to use an ODK container? #31

We need to try and test the same thing for dragon-ai agent and claude agent

[github-actions]

Issue Summary: Extend ODK Docker Image Usage to Claude and Dragon-AI Actions

Context

This issue builds on the successful experiments documented in #31, where GitHub Copilot was configured to use the Ontology Development Kit (ODK) Docker container. The goal is to replicate this pattern for the claude.yml and dragon-ai.yml GitHub Actions workflows.

Background from #31 (Copilot Implementation)

The discussion in #31 revealed several key learnings:

Key Findings:

• ✅ Proof of concept successful: @matentzn demonstrated that Copilot can successfully use the ODK container via the run.sh wrapper
• ✅ Works without explicit base image change: Instead of modifying the GitHub Action base image, the solution was to instruct the agent to use the run.sh wrapper script
• ✅ Instruction refinement critical: Early attempts showed agents would try to find ROBOT locally before using ODK. Adding prominent warnings to "read instructions first" ensured agents used the ODK wrapper from the start
• ⚠️ Known limitation: Copilot encounters issues when trying to open ontologies from PURLs through Docker (gets blocked)

Example Success:

• Copilot test in ai-curation-testing repo
• Instructions file used

Key instruction pattern:

NEVER run commands directly. ALL commands MUST use the ODK wrapper.

For ROBOT commands:
sh odk.sh robot COMMAND

For Make targets:
sh odk.sh make <target>

Current State: Claude and Dragon-AI Actions

Claude Action (.github/workflows/claude.yml):

• Currently uses ubuntu-latest base image
• Installs Python tools via uv sync
• Configured with MCP servers (OLS, sequential-thinking)
• No ODK Docker integration

Dragon-AI Action (.github/workflows/dragon-ai.yml):

• Currently uses ubuntu-latest base image
• Sets ROBOT_VERSION: v1.9.7 but doesn't show actual ROBOT installation
• Creates tools directory
• Installs Python tools via uv venv
• No ODK Docker integration

Why ODK Container Integration Matters

The ODK container provides:

1. Comprehensive tooling: Pre-installed ROBOT, OWLTools, and other ontology development utilities
2. Version consistency: Standardized tool versions across all AI agents
3. Reduced setup complexity: No need to install individual tools in each action
4. Domain expertise: Environment specifically optimized for ontology curation workflows

Implementation Approach

Based on the Copilot success, the recommended approach is:

1. Add Docker installation to both workflows (Docker is needed to run the ODK container)
2. Create/ensure ODK wrapper scripts (like run.sh or odk.sh) are available in the repository
3. Add custom instructions to guide Claude and Dragon-AI to use the ODK wrapper for all ontology-related commands
4. Test with simple ROBOT commands to verify the integration works

Related Issues

• Can we set up the copilot action to use an ODK container? #31 - Original Copilot/ODK discussion (open) ✅ Proof of concept successful
• Copilot is blocked from downloading OBO purls? #32 - PURL downloading issues with Copilot (open) ⚠️ Known limitation
• ODK image pulled _every single time_? #33 - ODK image caching optimization (open)
• Unify all the different ontology pipelines #40 - Unifying ontology pipelines across multiple repos (open) - This issue directly relates to standardization
• Create a list of exemplar repos that use ai4curation setup #47 - Gallery of example repos using ai4curation setup (open)

Next Steps

1. Document the Copilot pattern - The successful approach from Can we set up the copilot action to use an ODK container? #31 should be formally documented
2. Create template instructions - Develop standardized instructions for using ODK via wrapper scripts
3. Test with Claude - Modify claude.yml to use ODK and test on a sample ontology task
4. Test with Dragon-AI - Modify dragon-ai.yml to use ODK and test on a sample ontology task
5. Document differences - Note any agent-specific quirks or configuration needs
6. Create reusable template - As part of Unify all the different ontology pipelines #40's goal to unify pipelines, create a standardized approach

Technical Considerations

• Performance: Issue ODK image pulled _every single time_? #33 notes the ODK image is pulled every time, which impacts runtime
• Permissions: Both actions need docker command access
• Environment setup: May need to expose certain environment variables to the Docker container
• File mounting: Workspace files need to be accessible inside the container

Status: Ready for Implementation

The groundwork has been laid through the Copilot experiments. The pattern is proven and just needs to be adapted for Claude and Dragon-AI agents.