Oh My Opencode

Oh My OpenCode

Multi-agent orchestration plugin that transforms OpenCode into a full agent harness with specialized agents, background task execution, category-based model routing, and autonomous work modes.

Package: oh-my-opencode (install via bunx oh-my-opencode install) Repository: https://github.com/code-yeongyu/oh-my-opencode Schema: https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json

Prerequisites

1. OpenCode installed and configured (opencode --version should be 1.0.150+)

   curl -fsSL https://opencode.ai/install | bash
   # or: npm install -g opencode-ai
   # or: bun install -g opencode-ai
   

2. At least one LLM provider authenticated (opencode auth login)
3. Strongly recommended: Anthropic Claude Pro/Max subscription (Sisyphus uses Claude Opus 4.5)

Installation

Run the interactive installer:

bunx oh-my-opencode install

Non-interactive mode with provider flags:

bunx oh-my-opencode install --no-tui \
  --claude=<yes|no|max20> \
  --openai=<yes|no> \
  --gemini=<yes|no> \
  --copilot=<yes|no> \
  --opencode-zen=<yes|no> \
  --zai-coding-plan=<yes|no>

Verify:

opencode --version
cat ~/.config/opencode/opencode.json  # should contain "oh-my-opencode" in plugin array

Two Workflow Modes

Mode 1: Ultrawork (Quick Autonomous Work)

Include ultrawork or ulw in your prompt. That's it.

ulw add authentication to my Next.js app

The agent will automatically:

1. Explore your codebase to understand existing patterns
2. Research best practices via specialized background agents
3. Implement the feature following your conventions
4. Verify with diagnostics and tests
5. Keep working until 100% complete

Mode 2: Prometheus (Precise Planned Work)

For complex or critical tasks:

1. Press Tab → switches to Prometheus (Planner) mode
2. Describe your work → Prometheus interviews you, asking clarifying questions while researching your codebase
3. Confirm the plan → review generated plan in .sisyphus/plans/*.md
4. Run /start-work → Atlas orchestrator takes over:
   • Distributes tasks to specialized sub-agents
   • Verifies each task completion independently
   • Accumulates learnings across tasks
   • Tracks progress across sessions (resume anytime)

Critical rule: Do NOT use Atlas without /start-work. Prometheus and Atlas are a pair — always use them together.

Agents

All agents are enabled by default. Each has a default model and provider priority fallback chain.

Agent Invocation

Agents are invoked via delegate_task() or the --agent CLI flag — NOT with @ prefix.

// Invoke a specific agent
delegate_task(subagent_type="oracle", prompt="Review this architecture...")

// Invoke via category (routes to Sisyphus-Junior with category model)
delegate_task(category="visual-engineering", load_skills=["frontend-ui-ux"], prompt="...")

// Background execution (non-blocking)
delegate_task(subagent_type="explore", run_in_background=true, prompt="Find auth patterns...")

CLI:

opencode --agent oracle
opencode run --agent librarian "Explain how auth works in this codebase"

When to Use Which Agent

Categories

Categories route tasks to Sisyphus-Junior with domain-optimized models via delegate_task().

Category Usage

delegate_task(category="visual-engineering", load_skills=["frontend-ui-ux"], prompt="Create a dashboard component")
delegate_task(category="ultrabrain", load_skills=[], prompt="Design the payment processing flow")
delegate_task(category="quick", load_skills=["git-master"], prompt="Fix the typo in README.md")
delegate_task(category="deep", load_skills=[], prompt="Investigate and fix the memory leak in the worker pool")

Critical: Model Resolution Priority

Categories do NOT use their built-in defaults unless configured. Resolution order:

1. User-configured model (in oh-my-opencode.json) — highest priority
2. Category's built-in default (if category is in config)
3. System default model (from opencode.json) — fallback

To use optimal models, add categories to your config. See references/configuration.md.

Built-in Skills

Skills are injected into subagents via delegate_task(load_skills=[...]).

Slash Commands

Process Management

Background Agents

Fire multiple agents in parallel for exploration and research:

// Launch background agents (non-blocking)
delegate_task(subagent_type="explore", run_in_background=true, prompt="Find auth patterns in codebase")
delegate_task(subagent_type="librarian", run_in_background=true, prompt="Find JWT best practices")

// Collect results when needed
background_output(task_id="bg_abc123")

// Cancel all background tasks before final answer
background_cancel(all=true)

Concurrency Configuration

{
  "background_task": {
    "defaultConcurrency": 5,
    "staleTimeoutMs": 180000,
    "providerConcurrency": { "anthropic": 3, "google": 10 },
    "modelConcurrency": { "anthropic/claude-opus-4-5": 2 }
  }
}

Priority: modelConcurrency > providerConcurrency > defaultConcurrency

Tmux Integration

Run background agents in separate tmux panes for visual multi-agent execution:

{
  "tmux": {
    "enabled": true,
    "layout": "main-vertical",
    "main_pane_size": 60
  }
}

Requires running OpenCode in server mode inside a tmux session:

tmux new -s dev
opencode --port 4096

Layout options: main-vertical (default), main-horizontal, tiled, even-horizontal, even-vertical

Parallel Execution Patterns

Pattern 1: Explore + Librarian (Research Phase)

// Internal codebase search
delegate_task(subagent_type="explore", run_in_background=true, prompt="Find how auth middleware is implemented")
delegate_task(subagent_type="explore", run_in_background=true, prompt="Find error handling patterns in the API layer")

// External documentation search
delegate_task(subagent_type="librarian", run_in_background=true, prompt="Find official JWT documentation and security recommendations")

// Continue working immediately — collect results when needed

Pattern 2: Category-Based Delegation (Implementation Phase)

// Frontend work → visual-engineering category
delegate_task(category="visual-engineering", load_skills=["frontend-ui-ux"], prompt="Build the settings page")

// Quick fix → quick category
delegate_task(category="quick", load_skills=["git-master"], prompt="Create atomic commit for auth changes")

// Hard problem → ultrabrain category
delegate_task(category="ultrabrain", load_skills=[], prompt="Design the caching invalidation strategy")

Pattern 3: Session Continuity

// First delegation returns a session_id
result = delegate_task(category="quick", load_skills=["git-master"], prompt="Fix the type error")
// session_id: "ses_abc123"

// Follow-up uses session_id to preserve full context
delegate_task(session_id="ses_abc123", prompt="Also fix the related test file")

CLI Reference

Core Commands

opencode                           # Start TUI
opencode --port 4096               # Start TUI with server mode (for tmux integration)
opencode -c                        # Continue last session
opencode -s <session-id>           # Continue specific session
opencode --agent <agent-name>      # Start with specific agent
opencode -m provider/model         # Start with specific model

Non-Interactive Mode

opencode run "Explain closures in JavaScript"
opencode run --agent oracle "Review this architecture"
opencode run -m openai/gpt-5.2 "Complex reasoning task"
opencode run --format json "Query"    # Raw JSON output

Auth & Provider Management

opencode auth login                # Add/configure a provider
opencode auth list                 # List authenticated providers
opencode auth logout               # Remove a provider
opencode models                    # List all available models
opencode models anthropic          # List models for specific provider
opencode models --refresh          # Refresh models cache

Session Management

opencode session list              # List all sessions
opencode session list -n 10        # Last 10 sessions
opencode export <session-id>       # Export session as JSON
opencode import session.json       # Import session
opencode stats                     # Token usage and cost statistics
opencode stats --days 7            # Stats for last 7 days

Plugin & MCP Management

bunx oh-my-opencode install        # Install/configure oh-my-opencode
bunx oh-my-opencode doctor         # Diagnose configuration issues
opencode mcp list                  # List configured MCP servers
opencode mcp add                   # Add an MCP server

Server Mode

opencode serve --port 4096         # Headless server
opencode web --port 4096           # Server with web UI
opencode attach http://localhost:4096  # Attach TUI to running server

Built-in MCPs

Oh My OpenCode includes these MCP servers out of the box:

Hooks

All hooks are enabled by default. Disable specific hooks via disabled_hooks config:

Best Practices

Do

• Use ulw for quick autonomous tasks — just include the keyword in your prompt
• Use Prometheus + /start-work for complex projects — interview-based planning leads to better outcomes
• Configure categories for your providers — ensures optimal model selection instead of falling back to system default
• Fire explore/librarian agents in parallel — always use run_in_background=true
• Use session continuity — pass session_id for follow-up interactions with the same subagent
• Let the agent delegate — Sisyphus is an orchestrator, not a solo implementer
• Run bunx oh-my-opencode doctor to diagnose issues

Don't

• Don't use Atlas without /start-work — Atlas requires a Prometheus plan
• Don't manually specify models for every agent — the fallback chain handles this
• Don't disable todo-continuation-enforcer — it's what keeps the agent completing work
• Don't use Claude Haiku for Sisyphus — Opus 4.5 is strongly recommended
• Don't run explore/librarian synchronously — always background them

When to Use This Skill

• Installing or configuring oh-my-opencode
• Understanding agent roles and delegation patterns
• Troubleshooting model resolution or provider issues
• Setting up tmux integration for visual multi-agent execution
• Configuring categories for cost optimization
• Understanding the ultrawork vs Prometheus workflow choice

When NOT to Use This Skill

• General OpenCode usage unrelated to oh-my-opencode plugin features
• Provider authentication issues (use opencode auth directly)
• OpenCode core configuration (use OpenCode docs at https://opencode.ai/docs/)

Rules for the Agent

1. Package name is oh-my-opencode — NOT @anthropics/opencode or any other name
2. Use bunx (officially recommended) — not npx for oh-my-opencode CLI commands
3. Agent invocation uses --agent flag or delegate_task() — NOT @agent prefix
4. Never change model settings or disable features unless the user explicitly requests it
5. Sisyphus strongly recommends Opus 4.5 — using other models degrades the experience significantly
6. Categories do NOT use built-in defaults unless configured — always verify with bunx oh-my-opencode doctor --verbose
7. Prometheus and Atlas are always paired — never use Atlas without a Prometheus plan
8. Background agents should always use run_in_background=true — never block on exploration
9. Session IDs should be preserved and reused — saves 70%+ tokens on follow-ups
10. When using Ollama, set stream: false — required to avoid JSON parse errors

Auto-Notify on Completion

Background tasks automatically notify when complete via the background-notification hook. No polling needed — the system pushes completion events. Use background_output(task_id="...") only when you need to read the result.

Reference Documents

• Configuration Reference — Complete config with all agents, categories, provider chains, hooks, and options
• Troubleshooting Guide — Common issues and solutions