🏢 OPC (One-Person Company) — Multi-Agent 编排技能，让 AI 成为一人公司的 CEO。基于 OpenClaw 的 sessions_spawn 机制实现多 Agent 协作。
概要
— Turn OpenClaw into the CEO of a one-person company. — 让 OpenClaw 成为一人公司的 CEO。 OPC (One-Person Company) is an OpenClaw skill that turns complex tasks into multi-agent collaboration. You give the goal, the CEO (OpenClaw) breaks it down, hires the right agents, monitors progress, and delivers results. - — CEO reads your background, proposes a plan, waits for confirmation before executing - — Inject top expert mindsets (Kotler, Fowler, Jeff Dean...) into each agent role - — Author's own methodology framework, ready to inject into any agent role - — Survives context compaction; full lifecycle state in state.
README
以下は原文 README の抜粋です。原文を見る。
🏢 OPC — One-Person Company
Multi-Agent Orchestration Skill — Turn OpenClaw into the CEO of a one-person company.
多 Agent 编排技能 — 让 OpenClaw 成为一人公司的 CEO。
English
What is OPC?
OPC (One-Person Company) is an OpenClaw skill that turns complex tasks into multi-agent collaboration. You give the goal, the CEO (OpenClaw) breaks it down, hires the right agents, monitors progress, and delivers results.
You (Owner) → OpenClaw CEO → Sub-agents (Specialists)
Key Features
- Phase 0 Context Intake — CEO reads your background, proposes a plan, waits for confirmation before executing
- Built-in Persona Library — Inject top expert mindsets (Kotler, Fowler, Jeff Dean…) into each agent role
- LZW Advisor Persona — Author’s own methodology framework, ready to inject into any agent role
- Project State Persistence — Survives context compaction; full lifecycle state in
state.json - Checkpoint & Resume — Failed agents resume from last checkpoint, not from scratch
- Auto Root Cause Analysis — L1-L4 fault classification (platform / orchestration / skill / external)
- Aware Triggers — Declarative event-driven scheduling (cron / once / interval / on_message)
- Tool Discovery v2 — domain × capability tag system, precise skill recommendations
- User Model Evolution — CEO writes back learnings after each project; gets smarter over time
Quick Start
# Install
cd ~/.openclaw/skills
unzip agent-orchestration-v5.0.zip
# Trigger OPC by telling OpenClaw:
# "Help me build a complete [project]"
# "I need [multi-step task] done end-to-end"
Architecture
agent-orchestration-20260309-lzw/
├── SKILL.md ← Entry point (v5.0)
├── brain/ ← CEO decision layer
│ ├── core-flow.md ← 4-phase flow + Context Intake
│ ├── task-decomposition.md
│ ├── role-design.md
│ └── collaboration-patterns.md
├── engine/ ← Execution engine
│ ├── project_state.py ← Project lifecycle state
│ ├── trigger_engine.py ← Aware triggers
│ ├── tool_discovery.py ← Tool discovery (tag system v2)
│ └── diagnose_agent.py ← Auto root cause analysis
└── playbook/ ← Knowledge & templates
├── persona-priming.md ← Persona methodology + library index
├── personas/
│ └── lzw.md ← Built-in LZW Advisor Persona (v5.0)
├── templates/
└── scenarios/
Persona Priming
Two types of Personas, two design logics:
| Type | Example | How it works |
|---|---|---|
| Activation (public figures) | Kotler, Fowler, Jeff Dean | One-line reference activates LLM’s pre-trained knowledge |
| Injection (real individuals) | LZW (personas/lzw.md) |
Full methodology doc injected into agent context |
New paradigm: You can crystallize your own thinking framework into a Persona and inject it into your agent team.
Real-World Validation
推奨ツール
別のキーワードを試すか、フィルタを外してください。
インストール
npx skillfish add fozu-lzwpattern/opc-agent-orchestration