oh-my-opencode

What This Skill Does

oh-my-opencode is an advanced multi-agent orchestration plugin designed to transform OpenCode into a professional-grade autonomous agent harness. It implements a sophisticated workflow ecosystem where specialized AI agents collaborate to solve complex engineering tasks. By providing a structure for agent delegation, the skill facilitates high-level autonomy—ranging from simple code implementation to complex system architecture projects. The plugin integrates directly into the OpenCode lifecycle, enabling features such as category-based task routing, background task execution, tmux session integration, and persistent memory across development sessions via its unique planning architecture.

Installation

To begin using the plugin, ensure you have OpenCode version 1.0.150 or higher installed. You can install the skill by executing: bunx oh-my-opencode install. For automated environments, you may use the non-interactive mode: bunx oh-my-opencode install --no-tui --claude=max20 --openai=yes. Post-installation, verify that your configuration file at ~/.config/opencode/opencode.json includes the "oh-my-opencode" entry to ensure the plugin is correctly registered with the OpenCode runtime.

Use Cases

The primary use cases for oh-my-opencode involve complex software engineering workflows that require autonomous navigation and iterative problem-solving. Use the 'Ultrawork' mode for rapid feature development, where the agent autonomously explores, researches, and implements code. Utilize the 'Prometheus' mode for architectural tasks, major refactoring, or long-term projects where you require human-in-the-loop oversight, detailed task planning, and structured verification. It is ideal for teams or developers managing large repositories who need to delegate sub-tasks like bug fixing, documentation generation, or dependency management to specialized agents like Sisyphus, Atlas, or Librarian.

Example Prompts

1. "ulw add a JWT-based authentication system to the existing Next.js boilerplate and write the corresponding integration tests."
2. "I need to refactor the database layer for the entire project. Switch to Prometheus mode and interview me to generate a formal development plan."
3. "After the Prometheus plan is generated, execute /start-work so Atlas can orchestrate the task distribution to the sub-agents."

Tips & Limitations

For optimal performance, a high-tier LLM such as Claude 3.5 Sonnet or Opus 4.5 is strongly recommended, as the orchestration logic requires significant reasoning capabilities. Always review generated plans in .sisyphus/plans/ before executing /start-work to ensure alignment with your codebase standards. Note that background agents perform file operations autonomously, so ensure your git branch is clean before initiating long-running tasks. Always use Prometheus and Atlas as a combined pair to maintain state consistency across your project.