[Spike] Test OpenCode CLI Integration with GitLab Duo Self-Hosted

In 18.3, the Duo Agent Platform team announced the beta of CLI Agents in GitLab CI/CD. Of the CLI Agents offered, only OpenCode is available as an airgapped offering. We need to verify that the OpenCode AI CLI agent integration works properly with GitLab Self-Hosted instances. This test will validate the complete workflow from setup to execution.

Test Scope

Environment Requirements

• GitLab Self-Managed instance (latest)
• GitLab Ultimate subscription with GitLab Duo Enterprise add-on
• GitLab Duo Self-Hosted configuration with:
  • Self-hosted AI gateway
  • Supported local models
  • Active GitLab Runner for CI/CD execution

Prerequisites to Validate

• Beta and experimental features are enabled
  • GitLab Duo is properly configured and functional
  • Self-hosted models are properly configured
  • GitLab Runner is available and functional

CI/CD Variable Configuration

• Configure required environment variables in project settings:
  • ANTHROPIC_API_KEY (or OPENAI_API_KEY depending on model)
    • GITLAB_TOKEN_OPENCODE (service account token)
    • GITLAB_HOST (GitLab instance hostname)
  • Verify variables are masked and properly configured

Flow Configuration File

• Create .gitlab/duo/flows/opencode.yaml with proper OpenCode configuration
  • Validate the configuration includes:
    • OpenCode AI installation (npm install --global opencode-ai)
    • glab CLI installation and authentication
    • Git configuration
    • Proper authentication setup for OpenCode
    • Command execution with GitLab context variables

Flow Trigger Setup

• Create flow trigger in project settings
  • Configure trigger for "Mention" event type
  • Link to service account user and flow configuration file
  • Verify trigger appears in Automate > Flow triggers

End-to-End Workflow Testing

• Create test issue in project
  • Mention the OpenCode service account with a simple task (e.g., "@opencode-agent can you analyze the current repository structure?")
  • Verify pipeline starts and "Processing the request" message appears
  • Monitor CI/CD pipeline execution and logs
  • Confirm OpenCode AI successfully:
    • Connects to self-hosted models
    • Authenticates with GitLab API
    • Executes requested task
    • Provides response in issue comment

Error Handling and Edge Cases

• Test with invalid/expired API keys
  • Test with insufficient permissions
  • Test with complex multi-step tasks
  • Test with code modification requests
  • Verify proper error messages and logging

Definition of Done

• Test environment details are documented:
  • GitLab Version:
  • GitLab Duo Self-Hosted Version:
  • AI Gateway Version:
  • Model Backend: (e.g., vLLM)
  • Specific Model:
  • Runner Type: (Docker, Kubernetes, Shell)
• Self-Hosted customers can set up OpenCode for use within GitLab
• OpenCode CLI agent responds to mentions in issues/MRs
• Pipeline executes without errors
• Agent can read repository context and GitLab data via glab CLI
• Agent provides meaningful responses using self-hosted models
• All authentication flows work with self-hosted configuration
• Changes are properly committed and pushed when applicable
• Any potential issues are documented:
  • Authentication failures with self-hosted models
  • OpenCode configuration incompatibilities
  • CI/CD pipeline failures specific to self-hosted setup
  • Performance issues with local model inference
  • Network connectivity problems in isolated environments
• Documentation is updated (as appropriate):
  • GitLab Duo Self-Hosted setup documentation
  • OpenCode CLI agent configuration examples
  • Troubleshooting guides for self-hosted environments
  • Known limitations