Opencode Orchestrator

OpenCode Orchestrator

Trigger Workflow

Use the /use-opencode slash command to invoke this skill:

/use-opencode

This will guide you through constructing an optimal handoff to OpenCode.

When to Use OpenCode

1. Capabilities

Available Tools (within OpenCode)

• File read/write/search operations
• Shell command execution
• Code analysis and refactoring
• Git operations
• Web search and URL reading

2. Handoff Protocol

OpenCode does NOT share Antigravity's conversation context. You must explicitly provide:

1. File Paths: Absolute paths to all relevant files
2. Task Context: Background information and requirements
3. Constraints: Any specific approaches to use or avoid
4. Success Criteria: Clear definition of done

Command Structure

opencode run "YOUR_DETAILED_PROMPT"

Optional Flags

3. Prompt Construction

<context>
Project: [Project name and type]
Working Directory: [Absolute path]
Related Files: [List of relevant file paths]
</context>

<background>
[Any necessary context not obvious from files]
</background>

<task>
[Clear, specific task description]
</task>

<requirements>
- [Specific requirement 1]
- [Specific requirement 2]
</requirements>

<constraints>
- [What to avoid or preserve]
</constraints>

<success_criteria>
- [How to know the task is complete]
</success_criteria>

4. Model Selection

For Code-Heavy Tasks

opencode run -m github-copilot/claude-sonnet-4 "..."
# or
opencode run -m github-copilot/gemini-3-pro "..."

For Research/Documentation

opencode run -m github-copilot/gpt-4o "..."

Check Available Models

opencode models

Default Behavior

If no -m flag is provided, OpenCode uses the model configured in:

• ~/.opencode.json
• $XDG_CONFIG_HOME/opencode/.opencode.json
• ./.opencode.json (project-local)

5. Example Workflows

Code Review Delegation

opencode run "
<context>
Repository: /home/user/project
Branch: feature/new-auth
Compare against: main
</context>

<task>
Review the changes in the current branch compared to main.
Focus on security, performance, and code quality.
</task>

<requirements>
- Check for SQL injection vulnerabilities
- Identify N+1 query patterns
- Note any breaking API changes
</requirements>

<success_criteria>
- Written review summary in REVIEW.md
- Critical issues clearly flagged
- Suggested improvements provided
</success_criteria>
"

Research and Documentation

opencode run -m github-copilot/gpt-4o "
<context>
Project: /home/user/project
Tech Stack: Node.js, PostgreSQL, Redis
</context>

<task>
Research caching strategies for our API endpoints and create an implementation plan.
</task>

<requirements>
- Consider Redis caching patterns
- Evaluate cache invalidation strategies
- Account for our microservices architecture
</requirements>

<success_criteria>
- CACHING_PLAN.md created with recommendations
- Code examples for key patterns
- Migration steps outlined
</success_criteria>
"

6. Integration with Antigravity Workflow

When delegating from Antigravity to OpenCode:

1. Prepare Context: Gather all relevant file paths and requirements
2. Construct Prompt: Use the structured template above
3. Execute Handoff: Run via run_command tool
4. Monitor Output: Read OpenCode's logs and final result
5. Verify Results: Check file changes and run tests
6. Continue Work: Resume in Antigravity with updated codebase

Artifact Coordination

OpenCode may create its own files (plans, documentation). After handoff:

• Check for new files in the working directory
• Review any generated documentation
• Incorporate findings into Antigravity artifacts if relevant

When to Bring Work Back

Return work to Antigravity when:

• Browser testing is needed
• User interaction is required
• Results need integration with current conversation
• Visual verification is needed

7. Troubleshooting

Model-Specific Notes

• GitHub Copilot models: Require active Copilot subscription
• Anthropic models: Best for complex reasoning and code generation
• Gemini models: Strong at multi-file understanding
• OpenAI models: Effective for research and documentation

Debugging Commands

# Check OpenCode configuration
cat ~/.opencode.json

# List available models  
opencode models

# List configured agents
opencode agent list

# Check authentication status
opencode auth status

8. Advanced: Custom Agents

opencode agent create

This launches an interactive wizard to define:

• Agent name and description
• Custom system prompt
• Tool permissions
• Default model

Then invoke with:

opencode run -a my-custom-agent "Task description"

Example Use Cases

• Code Reviewer: Specialized prompts for security and performance review
• Documentation Writer: Focused on JSDoc/TSDoc generation
• Test Generator: Configured for your testing framework

Quick Reference

# Basic execution
opencode run "Your task description"

# With specific model
opencode run -m github-copilot/claude-sonnet-4 "Task"

# With custom agent
opencode run -a custom-agent "Task"

# Check available models
opencode models

# Check available agents
opencode agent list