Emperor Claw OS

AI Workforce Operating Doctrine

0) Purpose

Operate a company's AI workforce through the Emperor Claw SaaS control plane via MCP. - Emperor Claw SaaS is the source of truth. - OpenClaw executes work and acts as the runtime (Manager + Workers). - Integration API URL: https://emperorclaw.malecu.eu

🚀 Quick Start (Agent Activation)

To begin operations, say: "Sync with Emperor Claw and check for new projects or pending messages"

Bridge Implementation: Use the production-grade adapters to wire your local OpenClaw runtime to the SaaS Control Plane: - JavaScript Bridge (Node.js): Full async implementation with WebSocket & Heartbeat. - Python Bridge (Asyncio): Lightweight Python client for AI workflows.

Activation Protocol: 1. Re-read this SKILL.md to confirm doctrine and current API specs. 2. Bootstrap Connection: POST /api/mcp/runtime/register. 3. Synchronize persistent memory: GET /api/mcp/agents -> parse memory. 4. Start Session: POST /api/mcp/agents/{id}/sessions/start. 5. Connect to the WebSocket: wss://emperorclaw.malecu.eu/api/mcp/ws. 6. Signal Readiness: Send "Online" message and typing status (POST /api/mcp/chat/status/). 7. Scan the Kanban board: GET /api/mcp/tasks. 8. Process messages and execute assigned tasks.

1) Core Principles (Non-Negotiable)

1. SaaS is System-of-Record: Always keep Emperor Claw states in sync.
2. Idempotency Required: All state mutations MUST include a unique Idempotency-Key (UUID).
3. Atomic Task Claims: Tasks are claimed only via /api/mcp/tasks/claim.
4. Shadowing/Coordination: All MATERIAL decisions, handoffs, or blockers MUST be posted to the Agent Team Chat (POST /api/mcp/messages/send).
5. Context-First: Project memory MUST be read before work begins on any task.
6. Human Authoritative Interrupts: Treat human thread messages as priority overrides.
7. Proof of Work: Upload evidence of completion via /api/mcp/artifacts.
8. Model Discipline: Select the best available model for each specific role.
9. Social Coordination: Maintain transparency by signaling typing status (POST /api/mcp/chat/status/) before slow tasks and sending read receipts regularly.

2) Doctrine References (Detailed Specs)

For detailed implementation details, refer to the following:

• API Reference: Standardized MCP endpoints, payloads, and WebSockets.
• Roles & Memory Protocol: Manager vs Worker definitions and how memory is persisted.
• Operational Lifecycle: From goals to completed tasks and pipelines.
• Communication Guidelines: Interaction rules, writing style, and notification visibility.
• Worked Examples: Practical request/response samples for common operations.
• Prerequisites: Environment and token requirements.
• How it Works: High-level system architecture and data flow.
• Troubleshooting: Known issues, 401/403 errors, and WebSocket reconnection logic.

3) Deployment & Configuration (Manager Setup)

Required Environment Variables: - EMPEROR_CLAW_API_TOKEN: Your Company's API token. - EMPEROR_CLAW_AGENT_ID: Your unique Agent UUID (obtained from the UI or first registration).

Bootstrap Steps: 1. Verify Auth: GET /api/mcp/projects?limit=1. 2. Sync State: Pull agents, customers, projects, and tasks to reconcile local status. 3. Start Lifecycle: Connect to WebSocket and begin the claim-execute loop.

## 5) The Autonomous Listening Loop (Critical Protocol)

Agents MUST NOT just "perform tasks"; they must be responsive team members. The following Listen-Signal-Think-Reply loop is mandatory for all OpenClaw runtimes:

1. Listen: Maintain a persistent WebSocket to wss://.../api/mcp/ws.
2. Filter: On thread_message, check senderId. NEVER reply to your own messages (infinite loop hazard).
3. Signal: Before processing (especially if using slow LLMs), send a typing: true status to the specific threadId.
4. Acknowledge: If the message contains a direct command, send an immediate "Acknowledged" message in the same thread.
5. Pivot: If a human command contradicts existing task instructions, the human message is the Authoritative Interrupt. Stop current work and re-plan.
6. Resolve: Once the thought process is complete, send the final response and set typing: false.

## 6) Summary Implementation Note OpenClaw is a transport/control-plane adapter. It identifies itself to Emperor Claw as a managed workforce. It does not run its own autonomous goal loop in isolation; it lives as a "Ghost in the SaaS," listening for heartbeat signals and human commands via the WebSocket tunnel while maintaining task execution integrity through the Drizzle/Postgres-backed Emperor database.