oh-my-ssalsyphus

🔄 OpenCode port of oh-my-claudecode v3.0.11

Oh My Ssal Sisyphus - Multi-agent orchestration plugin that brings oh-my-claudecode features to OpenCode.

Like Sisyphus condemned to roll his boulder eternally, this plugin ensures your AI assistant never stops until the task is complete.

🎯 What is this?

This project ports the powerful features of oh-my-claudecode v3.0.11 (a Claude Code plugin) to the OpenCode platform.

| Original (Claude Code) | This Port (OpenCode) | |------------------------|----------------------| | oh-my-claudecode | oh-my-ssalsyphus | | Shell hooks + Node.js bridge | Native TypeScript plugin API | | Stop hook (proactive blocking) | session.idle event (reactive) |

Key Features Ported

• ✅ Ultrawork Mode (maximum intensity parallel execution)
• ✅ Ralph Loop (self-referential completion loop)
• ✅ TODO Continuation Enforcer
• ✅ PRD-based task tracking
• ✅ Notepad memory system (<remember> tags)
• ✅ Background agent orchestration
• ✅ Session state persistence

Features

Core Capabilities

• TODO Continuation Enforcer - Automatically reminds the assistant to complete pending tasks
• Ultrawork Mode - Maximum intensity parallel execution triggered by keywords
• Ralph Loop - Self-referential loop that continues until verified completion
• Background Agents - Run explore/librarian agents in parallel without blocking
• Agent Usage Reminder - Encourages using specialized agents over direct tool calls

Specialized Agents

Note: New agent names are available with backward compatibility for legacy names.

| Agent (New) | Alias (Legacy) | Model | Purpose | |-------------|----------------|-------|---------| | architect | oracle | Opus | Strategic architecture & debugging advisor (READ-ONLY) | | architect-medium | oracle-medium | Sonnet | Balanced architectural analysis (READ-ONLY) | | architect-low | oracle-low | Haiku | Fast architectural checks (READ-ONLY) | | researcher | librarian | Sonnet | Documentation & reference researcher | | researcher-low | librarian-low | Haiku | Quick documentation lookup | | explore | - | Haiku | Fast codebase search | | explore-medium | - | Sonnet | Deeper codebase analysis | | designer | frontend-engineer | Sonnet | UI/UX designer-developer | | designer-low | frontend-engineer-low | Haiku | Fast UI changes | | designer-high | frontend-engineer-high | Opus | Complex UI architecture | | writer | document-writer | Haiku | Technical documentation | | executor | sisyphus-junior | Sonnet | Focused task executor | | executor-low | sisyphus-junior-low | Haiku | Simple task execution | | executor-high | sisyphus-junior-high | Opus | Complex multi-file tasks | | planner | prometheus | Opus | Strategic planning | | analyst | metis | Opus | Pre-planning analysis | | critic | momus | Opus | Plan review | | vision | multimodal-looker | Sonnet | Visual/media analysis | | qa-tester | - | Sonnet | Interactive CLI testing |

Installation

# Using npm
npm install omo-omcs

# Using bun
bun add omo-omcs

# Using pnpm
pnpm add omo-omcs

Then add to your OpenCode configuration:

{
  "plugins": ["omo-omcs"]
}

Usage

Ultrawork Mode

Trigger maximum performance mode by including keywords in your prompt:

ultrawork: Implement the authentication system with tests

or

ulw refactor the database layer

Keywords: ultrawork, ulw, uw

Slash Commands

| Command | Description | |---------|-------------| | /ultrawork <task> | Maximum intensity parallel execution | | /ralph-loop <task> | Completion guarantee loop | | /ultrawork-ralph <task> | Combined max intensity + completion guarantee | | /ultraqa <goal> | QA cycling workflow | | /ralplan <task> | Iterative planning with Planner/Architect/Critic | | /plan <task> | Start planning session | | /review <plan> | Review plan with Critic | | /doctor | Diagnose installation issues | | /cancel-ralph | Cancel Ralph Loop | | /cancel-ultraqa | Cancel UltraQA |

Ralph Loop

Start a self-referential loop that continues until the task is verified complete:

/ralph-loop "Implement user registration with email verification"

Cancel with:

/cancel-ralph

Background Agents

The plugin provides tools for running agents in the background:

// Run explore agent in background
background_task(agent="explore", prompt="Find all authentication-related files")

// Run librarian for documentation research
background_task(agent="librarian", prompt="Find JWT best practices documentation")

// Get results when ready
background_output(task_id="...")

// Cancel all background tasks
background_cancel(all=true)

Call Agents Directly

// Synchronous call
call_omo_agent(
  subagent_type="oracle",
  prompt="Review this architecture decision...",
  run_in_background=false
)

// Async call
call_omo_agent(
  subagent_type="explore",
  prompt="Find all React components",
  run_in_background=true
)

Configuration

Create .opencode/omo-omcs.json in your project:

{
  "disabled_hooks": [],
  "background_task": {
    "max_concurrent": 5,
    "timeout_ms": 300000
  },
  "ralph_loop": {
    "max_iterations": 50,
    "idle_timeout_ms": 30000
  },
  "todo_continuation": {
    "idle_threshold_ms": 15000,
    "reminder_interval_ms": 60000
  },
  "ultrawork": {
    "keywords": ["ultrawork", "ulw", "uw"],
    "auto_parallel": true
  }
}

Configuration Options

| Option | Description | Default | |--------|-------------|---------| | disabled_hooks | Hooks to disable | [] | | background_task.max_concurrent | Max parallel background tasks | 5 | | background_task.timeout_ms | Background task timeout | 300000 | | ralph_loop.max_iterations | Max loop iterations | 50 | | ralph_loop.idle_timeout_ms | Idle time before continuation | 30000 | | todo_continuation.idle_threshold_ms | Idle threshold for TODO check | 15000 | | ultrawork.keywords | Trigger keywords | ["ultrawork", "ulw", "uw"] |

Model Provider Configuration

By default, oh-my-ssalsyphus uses GitHub Copilot Claude 4.5 models for the three tiers:

| Tier | Default Model | |------|---------------| | haiku | github-copilot/claude-haiku-4.5 | | sonnet | github-copilot/claude-sonnet-4.5 | | opus | github-copilot/claude-opus-4.5 |

To use other providers like Google or OpenAI, configure model_mapping.tierDefaults:

{
  "model_mapping": {
    "tierDefaults": {
      "haiku": "google/gemini-3-flash",
      "sonnet": "google/gemini-3-pro-low",
      "opus": "openai/gpt-5.2"
    }
  }
}

Model Format: Always use provider/model-name format (e.g., github-copilot/claude-sonnet-4.5, google/gemini-3-flash)

Tier Semantics: | Tier | Purpose | Typical Use | |------|---------|-------------| | haiku | Fast, cheap | explore, simple lookups, quick checks | | sonnet | Balanced | executor, designer, most agents | | opus | Most capable | architect, planner, complex reasoning |

Partial Configuration: You only need to specify tiers you want to override. Unspecified tiers fall back to Copilot Claude defaults.

Per-Agent Override: Override specific agents with either a tier or concrete model:

{
  "agents": {
    "architect": { "model": "openai/gpt-5.2" },
    "explore": { "tier": "sonnet" },
    "executor": { "model": "google/gemini-3-pro-low", "temperature": 0.3 }
  }
}

Debug Logging: Enable model resolution logging:

{
  "model_mapping": {
    "debugLogging": true
  }
}

Disabling Hooks

{
  "disabled_hooks": [
    "agent-usage-reminder",
    "keyword-detector"
  ]
}

Available hooks:

• todo-continuation-enforcer
• keyword-detector
• ralph-loop
• session-recovery
• agent-usage-reminder

How It Works

TODO Continuation Enforcer

When the assistant has incomplete TODO items and becomes idle:

1. Plugin detects idle state (no tool calls for configured threshold)
2. Checks if pending/in_progress TODOs exist
3. Injects continuation prompt:

[SYSTEM REMINDER - TODO CONTINUATION]
Incomplete tasks remain in your todo list. Continue working on the next pending task.
Proceed without asking for permission. Mark each task complete when finished.
Do not stop until all tasks are done.

Ralph Loop

Self-referential execution loop:

1. User invokes /ralph-loop "task description"
2. Plugin creates PRD (Product Requirements Document) if not exists
3. Monitors for completion signals (<promise>DONE</promise>)
4. Re-injects task prompt on idle until completion
5. Tracks iterations to prevent infinite loops

Ultrawork Mode

When triggered:

1. Injects high-intensity system prompt
2. Encourages parallel execution
3. Minimizes confirmation requests
4. Maximizes throughput

Development

# Clone the repo
git clone https://github.com/devswha/oh-my-ssalsyphus.git
cd oh-my-ssalsyphus

# Install dependencies
bun install

# Build
bun run build

# Type check
bun run typecheck

# Run tests
bun test

Philosophy

Named after Sisyphus from Greek mythology - condemned to roll a boulder up a hill for eternity. Like Sisyphus, this plugin embodies persistence:

• Tasks are tracked until completion
• The assistant cannot simply "give up"
• Work continues until the boulder reaches the summit

"The struggle itself toward the heights is enough to fill a man's heart. One must imagine Sisyphus happy." - Albert Camus

Credits

This project is an OpenCode port of oh-my-claudecode by @Yeachan-Heo.

Special thanks to:

• oh-my-claudecode - The original Claude Code plugin this project ports
• OpenCode - The target platform

License

MIT