Oh My OpenCode
A “battery-included” plugin for OpenCode that transforms it into a powerful multi-agent orchestration system. Think of it as Ubuntu to OpenCode’s Debian.
What’s Included
Agents (Your AI Team)
| Agent | Model | Purpose |
|---|---|---|
| Sisyphus | Claude Opus 4.5 | Default orchestrator with 32k thinking budget. Plans, delegates, executes with parallel execution |
| oracle | GPT-5.2 | Architecture, code review, strategy (inspired by AmpCode) |
| librarian | Claude Sonnet 4.5 | Multi-repo analysis, doc lookup, implementation examples |
| explore | Grok | Fast codebase exploration and pattern matching (free & fast) |
| frontend-ui-ux-engineer | Gemini 3 Pro | UI/UX specialist for beautiful interfaces |
| document-writer | Gemini 3 Flash | Technical writing expert |
| multimodal-looker | Gemini 3 Flash | Analyzes PDFs, images, diagrams |
Tools & Features
Context Management
- Directory AGENTS.md/README.md Injector - Auto-injects context files when reading, walking from file to project root
- Conditional Rules Injector - Injects rules from
.claude/rules/when conditions match (supports globs, alwaysApply) - Context Window Monitor - Implements “Context Window Anxiety Management” (warns at 70%+ usage)
Built-in MCPs
- context7 - Official documentation lookup
- websearch_exa - Real-time web search
- grep_app - Ultra-fast code search across public GitHub repos
LSP/AST Tools
- Full LSP integration with refactoring tools (rename, code actions)
- AST-Grep for advanced code searching
Smart Automation
- Keyword Detector - Activates modes based on prompts:
ultrawork/ulw- Maximum performance with parallel agentssearch/find- Maximized search with parallel explore + librariananalyze/investigate- Deep multi-phase expert consultation
- Think Mode - Auto-detects when extended thinking is needed (“ultrathink”)
- Todo Continuation Enforcer - Forces agents to complete all TODOs
- Comment Checker - Reduces excessive LLM comments in code
Stability Features
- Anthropic Auto Compact - Auto-summarizes when hitting token limits
- Session Recovery - Recovers from errors (missing tool results, thinking block issues)
- Empty Task Response Detector - Catches when Task tool returns nothing
- Grep Output Truncator - Dynamically truncates based on context window
Notifications
- Background task completion notifications
- Session idle notifications (macOS, Linux, Windows)
- Auto-update checker
AmpCode/Claude Code Compatibility
look_attool ported from AmpCode- Claude Code compatibility layer
Installation
# Interactive installer
npx oh-my-opencode
# Or let an LLM agent install it
# Paste the GitHub URL into a fresh opencode session Configuration
Config locations (priority order):
.opencode/oh-my-opencode.json(project)~/.config/opencode/oh-my-opencode.json(user)
Customize Agents
{
"agents": {
"oracle": {
"model": "openai/gpt-4o",
"temperature": 0.7,
"disable": false
}
}
} Each agent supports: model, temperature, top_p, prompt, tools, disable, description, mode, color, permission
Disable Agents
{
"disabled_agents": ["frontend-ui-ux-engineer", "document-writer"]
} LSP Configuration
{
"lsp": {
"typescript": {
"command": "typescript-language-server --stdio",
"extensions": [".ts", ".tsx"]
}
}
} Key Differentiators
- Multi-model orchestration - Uses the best model for each task
- Aggressive parallel execution - Background agents work simultaneously
- No token bloat - Efficient context management
- Production-tested - Author spent $24,000 in tokens refining it
- Supports existing subscriptions - Works with ChatGPT, Claude, Gemini subscriptions