oh-my-openclaw

Oh-My-OpenClaw (OmOC)

Agent orchestration skill that brings structured planning, execution, and knowledge accumulation to OpenClaw.

Overview

Oh-My-OpenClaw ports the proven patterns from oh-my-opencode into OpenClaw-native constructs:

• 3-Layer Agent Architecture: Planning (Prometheus/Metis/Momus) -> Orchestration (Atlas) -> Execution (Sisyphus-Junior/Hephaestus/Oracle/Explore/Librarian)
• Native Multi-Agent: Uses OpenClaw sessions_spawn for real sub-agent sessions (not just role-switching)
• Category System: Intent-based model routing (quick/deep/ultrabrain/visual-engineering)
• Wisdom Accumulation: File-based notepad system for persistent learnings across sessions
• Ultrawork Mode: One-command full automation from planning to verified completion
• Todo Enforcer: System prompt injection ensuring forced task completion
• Tool Restriction: Native OpenClaw agents.list[].tools.profile/allow/deny for per-agent access control

Installation

openclaw plugins install @happycastle/oh-my-openclaw

Skills, hooks, and tools are registered automatically. Run openclaw omoc-setup to inject agent configs.

Agent Tool Restrictions

Inject agent configs into your OpenClaw config for sub-agent spawning:

openclaw omoc-setup
# Use --force to overwrite existing, --dry-run to preview

Verify Installation

The skill should appear in OpenClaw's available skills list. Test by asking:

"Read the oh-my-openclaw skill and tell me what it does"

Trigger

This skill activates when:

• User invokes /ultrawork, /plan, or /start_work commands
• User requests complex multi-step task planning
• User asks for agent orchestration or delegation

Architecture

Layer 1: Planning

Layer 2: Orchestration

Layer 3: Workers

Category-to-Model Mapping

Categories map user intent to optimal model selection:

{
  "quick": "claude-sonnet-4-6",
  "deep": "claude-opus-4-6-thinking",
  "ultrabrain": "gpt-5.3-codex",
  "visual-engineering": "claude-opus-4-6-thinking"
}

Workflows

/ultrawork - Full Automation Loop

1. Prometheus creates a strategic plan via user interview
2. Momus reviews and critiques the plan
3. Atlas breaks plan into executable tasks
4. Workers execute tasks with Todo tracking
5. Atlas verifies completion of each task
6. Loop continues until all tasks are done

/plan - Planning Only

1. Prometheus interviews user about the task
2. Creates a phased plan saved to workspace/plans/
3. Momus reviews the plan
4. Returns refined plan for user approval

/start_work - Execute Existing Plan

1. Reads plan from workspace/plans/
2. Atlas distributes tasks to appropriate workers
3. Workers execute with Todo tracking
4. Verification loop until completion

tmux/OmO Delegation (Skills)

Coding delegation and multi-session orchestration are handled by dedicated skills (loaded automatically):

Wisdom Accumulation

The notepad system persists learnings across sessions:

workspace/notepads/
  learnings.md    - Technical discoveries and patterns
  decisions.md    - Architecture and design decisions made
  issues.md       - Known issues and workarounds
  preferences.md  - User preferences and conventions

How It Works

• Workers automatically append discoveries to relevant notepads
• Planning agents read notepads before creating plans
• Notepads survive across sessions (file-based persistence)
• Each entry is timestamped and tagged with source context

Todo Enforcer

System prompt injection that ensures task completion:

[SYSTEM DIRECTIVE: OH-MY-OPENCLAW - TODO CONTINUATION]
You MUST continue working on incomplete todos.
- Do NOT stop until all tasks are marked complete
- Do NOT ask for permission to continue
- Mark each task complete immediately when finished
- If blocked, document the blocker and move to next task

Ralph Loop

Self-referential completion mechanism:

1. After completing a task batch, agent reviews remaining todos
2. If incomplete items exist, agent continues without user intervention
3. Loop terminates only when all todos are complete or explicitly cancelled
4. Maximum iterations configurable (default: 10)

Built-in Skills

Skills inject specialized knowledge and workflows into agents. Load them via load_skills when delegating tasks.

Category + Skill Combos

File Structure

oh-my-openclaw/
  SKILL.md              # This file - main skill instructions
  README.md             # Project documentation
  LICENSE               # MIT license
  CHANGELOG.md          # Version history
  config/
    categories.json     # Category-to-model mapping + tool restrictions + skill triggers
  docs/                 # Documentation files
  plugin/
    agents/
      prometheus.md       # Strategic planner agent profile
      metis.md            # Pre-planning consultant (intent classification + anti-slop directives)
      momus.md            # Practical plan reviewer (critical blocker checks)
      atlas.md            # Task orchestrator agent profile
      sisyphus-junior.md  # Primary worker agent profile
      hephaestus.md       # Autonomous deep worker for complex execution
      oracle.md           # Architect/debugger agent profile
      librarian.md        # Documentation specialist agent profile
      explore.md          # Search specialist agent profile
      multimodal-looker.md # Visual analysis agent profile
    skills/
      git-master.md       # Git expert skill (commits, rebase, history)
      frontend-ui-ux.md   # Design-first UI development skill
      comment-checker.md  # Anti-AI-slop code quality skill
      gemini-look-at.md   # Gemini CLI multimodal analysis (PDF/image/video)
      web-search.md       # Web search integration (Exa/Context7/grep.app MCP)
      opencode-controller.md  # OpenCode/OmO delegation patterns
      tmux.md             # tmux session orchestration patterns
      tmux-agents.md      # Agent spawning/monitoring in tmux
      workflow-tool-patterns.md # OmO→OpenClaw tool mapping
      workflow-auto-rescue.md   # Checkpoint-based recovery
    workflows/
      ultrawork.md        # Full automation workflow
      plan.md             # Planning-only workflow
      start-work.md       # Execute existing plan workflow
    src/                # TypeScript plugin source
    dist/               # Compiled plugin

Usage Examples

Quick Task

User: Fix the type error in auth.ts
Agent: [Uses Sisyphus-Junior directly - category: quick]

Complex Feature

User: /ultrawork Add user authentication with OAuth2
Agent: [Prometheus plans -> Momus reviews -> Atlas distributes -> Workers execute]

Research Task

User: /plan Research the best approach for real-time notifications
Agent: [Prometheus + Librarian + Oracle collaborate on research plan]