# How to Configure and Deploy the OpenMAIC Skill with OpenClaw Workbench Integration > Learn to configure and deploy the OpenMAIC Skill with OpenClaw workbench integration. Follow our guide for seamless setup and get started quickly. - Repository: [MAIC/OpenMAIC](https://github.com/THU-MAIC/OpenMAIC) - Tags: how-to-guide - Published: 2026-09-13 --- **The OpenMAIC Skill deploys to OpenClaw through a standard SKILL.md package located in `skills/openmaic/`, requiring configuration in `~/.openclaw/openclaw.json` and a running OpenMAIC server instance with valid LLM provider keys.** The **THU-MAIC/OpenMAIC** repository provides a ready-to-use skill package that integrates with the OpenClaw agent workbench, enabling natural language classroom generation workflows. This guide walks through the complete OpenMAIC Skill OpenClaw workbench integration, from installation to asynchronous classroom deployment. ## Understanding the OpenMAIC Skill Architecture The skill follows the standard **SKILL.md** format recognized by OpenClaw and other compliant workbenches. According to the source code analysis, the integration operates through four logical phases defined in [`skills/openmaic/SKILL.md`](https://github.com/THU-MAIC/OpenMAIC/blob/main/skills/openmaic/SKILL.md): - **Clone**: Detects whether the OpenMAIC repository exists locally and prompts for installation if missing. - **Startup**: Guides server initialization via `pnpm dev`, production builds, or Docker, verifying that the instance is reachable. - **Provider Keys**: Surfaces required environment variables from `.env.local` and suggests optimal LLM providers. - **Generation**: Submits async jobs to the `/api/generate-classroom` endpoint, polls for completion, and returns classroom URLs. The runtime logic resides in `lib/server/agent-runtime/`, which handles session management, material processing, and job polling during the generation phase. ## Installing the OpenMAIC Skill in OpenClaw Install the skill through the OpenClaw CLI to register it with your workbench. ```bash clawhub install openmaic ``` If you are using an alternative workbench (such as Codex, DeepSeek, or WorkBuddy), manually import the `skills/openmaic/` directory or its compressed archive into your workbench's skill repository. The package contains the [`SKILL.md`](https://github.com/THU-MAIC/OpenMAIC/blob/main/SKILL.md) router and the `references/` folder containing inline help fragments. ## Configuring the Skill in OpenClaw OpenClaw reads per-skill configuration from `~/.openclaw/openclaw.json`. You must add an **openmaic** entry specifying either hosted or self-hosted mode. ### Hosted Mode Configuration For the managed service at `https://open.maic.chat/`, provide your access code: ```json { "skills": { "entries": { "openmaic": { "config": { "accessCode": "sk-YOUR-ACCESS-CODE" } } } } } ``` ### Self-Hosted Mode Configuration For local development or private deployments, specify the repository path and server URL: ```json { "skills": { "entries": { "openmaic": { "config": { "repoDir": "/path/to/OpenMAIC", "url": "http://localhost:3000" } } } } } ``` Place this file at `~/.openclaw/openclaw.json` as documented in the README under the **OpenClaw Integration** section. ## Setting Up the OpenMAIC Server The skill requires a running OpenMAIC server instance. Deploy this through Node.js or Docker before invoking the skill. ### Clone and Install Dependencies ```bash git clone https://github.com/THU-MAIC/OpenMAIC.git cd OpenMAIC pnpm install ``` ### Configure Environment Variables Copy the template and configure at least one LLM provider: ```bash cp .env.example .env.local ``` Edit `.env.local` to include your provider credentials: ```env # Required: At least one LLM provider OPENAI_API_KEY=sk-xxxxxxxxxxxx DEFAULT_MODEL=openai:gpt-5.5 # Optional: Azure OpenAI configuration # AZURE_OPENAI_API_KEY=your-azure-key # AZURE_OPENAI_BASE_URL=https://YOUR-RESOURCE.openai.azure.com/openai # Optional: Local Lemonade provider LEMONADE_BASE_URL=http://localhost:13305/v1 ``` The repository's `.env.example` contains the full list of supported providers and variables. ### Start the Server Run the development server: ```bash pnpm dev ``` For production deployments: ```bash pnpm build && pnpm start ``` The server listens on `http://localhost:3000` by default. Alternatively, use Docker for persistent deployments: ```yaml services: openmaic: build: . env_file: .env.local ports: - "3000:3000" postgres: image: postgres:16-alpine environment: POSTGRES_USER: openmaic POSTGRES_PASSWORD: openmaic-dev POSTGRES_DB: openmaic volumes: - pgdata:/var/lib/postgresql/data volumes: pgdata: ``` ## Using the OpenMAIC Skill Integration With the server running and skill configured, invoke classroom generation through natural language commands in any OpenClaw chat window: ``` teach me quantum physics ``` OpenClaw executes the following workflow: 1. **Configuration Verification**: Validates `accessCode` for hosted mode or `repoDir`/`url` for self-hosted mode. 2. **Endpoint Communication**: POSTs to `/api/generate-classroom` with any attached material files. 3. **Job Polling**: Monitors the async generation job status through the agent runtime. 4. **Result Delivery**: Returns a clickable link to the generated classroom (e.g., `http://localhost:3000/classroom/abc123`). Each step requires explicit user confirmation, preventing unauthorized automation. ## Extending the Skill for Custom Workflows Advanced users can modify the skill behavior by editing files under `skills/openmaic/`: - **[`SKILL.md`](https://github.com/THU-MAIC/OpenMAIC/blob/main/SKILL.md)**: Defines the SOP flow and routing logic between clone, startup, and generation phases. - **`references/`**: Contains Markdown snippets rendered as contextual help during each step. After modification, re-import the directory into your workbench to deploy custom classroom generation workflows. ## Summary - The **OpenMAIC Skill** package resides in [`skills/openmaic/SKILL.md`](https://github.com/THU-MAIC/OpenMAIC/blob/main/skills/openmaic/SKILL.md) and follows the standard skill format. - Configure OpenClaw through `~/.openclaw/openclaw.json` using either `accessCode` (hosted) or `repoDir`/`url` (self-hosted). - The OpenMAIC server requires `pnpm install`, a populated `.env.local` file, and at least one LLM provider key. - Natural language commands trigger the `/api/generate-classroom` endpoint via the `lib/server/agent-runtime/` subsystem. - Docker and secondary development options support production deployments and custom SOP modifications. ## Frequently Asked Questions ### What file format does the OpenMAIC Skill use for workbench integration? The skill uses the **SKILL.md** format located at [`skills/openmaic/SKILL.md`](https://github.com/THU-MAIC/OpenMAIC/blob/main/skills/openmaic/SKILL.md). This Markdown-based specification defines the standard operating procedure (SOP) flow that OpenClaw and other compliant workbenches parse to execute the clone, startup, provider configuration, and generation phases. ### Where does OpenClaw store the OpenMAIC Skill configuration? OpenClaw stores skill-specific configuration in the `~/.openclaw/openclaw.json` file. The **openmaic** entry within the `skills.entries` object accepts `accessCode` for hosted mode or `repoDir` and `url` for self-hosted deployments. ### Which LLM providers can I use with the OpenMAIC server? The OpenMAIC server supports **OpenAI**, **Azure OpenAI**, and local providers like **Lemonade**. Configure these in the `.env.local` file by setting `OPENAI_API_KEY`, `AZURE_OPENAI_API_KEY` with `AZURE_OPENAI_BASE_URL`, or `LEMONADE_BASE_URL` respectively. The `DEFAULT_MODEL` variable specifies which provider to use by default. ### How does the skill handle classroom generation asynchronously? When you submit a generation request, the skill calls the `/api/generate-classroom` endpoint and creates a durable job managed by `lib/server/agent-runtime/`. The skill polls this job status until completion, then returns the final classroom URL, ensuring reliable handling of long-running generation tasks without blocking the workbench interface.