Oh My OpenCode (5 SUPER Agent/MCP/Prompt Config): This makes OPENCODE - A BEAST! REALLY GOOD Agents

Oh My Open Code: A Detailed Overview

Key Concepts:

• Oh My Open Code (OMO): A configuration framework for the Open Code CLI, designed to orchestrate multiple AI models for coding tasks.
• Cisphus: The main agent in OMO, acting as an “engineering manager” and delegating tasks to specialized sub-agents. Typically runs on Anthropic’s Claude Opus 4.5.
• Sub-agents: Specialized AI agents running on different models (Gemini 3 Pro, Claude Sonnet 4.5, GPT 5.2) focused on specific tasks like UI development, research, or architectural review.
• LSP (Language Server Protocol): A protocol enabling agents to perform diagnostics and fix syntax errors in real-time.
• MCP (Model Context Protocol): Configurations defining how agents interact with specific models.
• Todo Continuation Enforcer: A mechanism preventing agents from prematurely stopping tasks, ensuring complete file generation.
• Anti-gravity Rate Limits: Google Gemini’s rate limiting system, potentially more generous than standard limits.

---

1. The Problem with Current AI Coding Tools & OMO’s Solution

The speaker highlights a key limitation of current AI coding agents like Cursor, Windsurf, and standard CLI agents: they typically rely on a “monomodel workflow.” This means users are locked into the strengths and weaknesses of a single AI model. For example, Gemini 3 Pro excels at front-end development but struggles with complex back-end logic, while Claude Opus 4.5 is strong overall but expensive for simple tasks. GPT 5.2 is good for architecture but can be inflexible with formatting. Manually switching between models to leverage their individual strengths is cumbersome.

Oh My Open Code (OMO) addresses this by acting as an orchestrator. It doesn’t replace the core Open Code CLI but enhances it with “ridiculous amount of steroids, plugins, and configurations.” OMO treats the main agent, Cisphus, not as a coder, but as an engineering manager, delegating tasks to specialized sub-agents running on different models. This allows for a more efficient and effective coding process by leveraging the best tool for each specific job.

2. Agent Roles and Model Allocation

OMO utilizes a team of specialized agents:

• Cisphus (Claude Opus 4.5): The central agent responsible for planning, task delegation, and overall project management.
• Front-end UI/UX Engineer (Gemini 3 Pro): Handles UI component creation and styling.
• Librarian (Claude Sonnet 4.5): Focuses on research, documentation lookup, and gathering implementation details (e.g., TMDB API documentation).
• Oracle (GPT 5.2): Provides sanity checks on architectural decisions.

These agents operate asynchronously in the background, enabling parallel processing. While one agent is building React components, another can be researching the database schema. This parallel execution is a significant advantage over single-model agents.

3. Addressing Agent “Laziness” & Improving Code Quality

The speaker points out a common issue with current AI agents: they often write incomplete code and leave comments indicating unfinished tasks. OMO tackles this with a “todo continuation enforcer” that forces the model to complete the file before stopping.

Furthermore, OMO integrates with LSP (Language Server Protocol), allowing agents to run diagnostics and fix syntax errors before presenting the code to the user. This proactive error correction significantly improves code quality and reduces the need for manual debugging.

4. Installation and Configuration

Installing OMO involves running the bunx oh my open code install command. The installer prompts the user for API access or subscriptions to services like Claude, Gemini, and GPT. While using APIs is possible, the speaker recommends subscriptions for optimal performance. The installation process configures MCPs (Model Context Protocols) and context web search, some of which don’t require additional APIs. Authentication with subscribed services is required. The speaker notes that Google Gemini utilizes “anti-gravity rate limits,” which can be more generous.

5. Demonstration: Building a Movie Tracker App

The speaker demonstrates OMO by prompting it to build a movie tracker app using the TMDB API with a minimalist aesthetic and a git-style contribution graph.

Step-by-Step Process:

1. Model Selection: The speaker selects Claude Opus 4.5 as the Cisphus model using the /models command.
2. Task Breakdown: Cisphus immediately breaks down the request into TMDB API integration, UI/UX design, and the contribution graph.
3. Delegation: Cisphus delegates tasks:
   • The Librarian agent researches the TMDB API (authentication, endpoints, rate limits). This runs in the background (run in background = true).
   • The Front-end UI/UX Engineer agent is tasked with styling the UI.
4. Code Generation & Refactoring: Cisphus begins generating TypeScript types and then incorporates the research findings from the Librarian agent. It also identifies and fixes unnecessary comments and lint issues thanks to LSP integration.
5. Directory Structure Creation: OMO creates the necessary directory structure using mkdir commands.
6. Component Creation: Components like movie search, movie card, watched movies list, and contribution graph are created.
7. Error Correction: When a module import error occurs, Cisphus leverages LSP to identify and correct the import path. It also addresses a potential race condition causing an unused variable error.
8. Configuration Updates: OMO updates app/layout.tsx and app/page.tsx to integrate the new components. It also configures next.config.ts to allow TMDB images.
9. UI Polish: The Front-end UI/UX Engineer agent (Gemini 3 Pro) handles the visual styling.
10. Build Verification: Cisphus initiates an npm run build command to ensure the application compiles successfully.

6. Cost Considerations and Future Potential

The speaker acknowledges that OMO isn’t a “magical free tool.” Orchestrating multiple models requires access to APIs or subscriptions from Anthropic, Google, and OpenAI, potentially leading to higher costs than using a single, cheaper model. However, OMO aims to mitigate this by offloading simple tasks to less expensive models like Haiku or Flash.

The speaker suggests the possibility of integrating other models like GLM and offers to create a tutorial on editing model configurations.

7. Conclusion & Key Takeaways

The speaker concludes that OMO is a “pretty cool” tool with significant potential for productivity gains. The dynamic delegation to specialized agents, parallel processing, LSP integration, and todo continuation enforcer are all major advantages. While cost remains a factor, the ability to leverage the strengths of multiple AI models offers a compelling alternative to traditional single-model agents. The demonstration showcased OMO’s ability to not only generate code but also to self-correct errors and apply best practices, making it a powerful tool for experienced developers.

Quote: “It’s like having a miniature dev team.” – Describing the benefits of OMO’s agent delegation.