What does this skill do, and when should you use it?
This is the official OpenClaw skill for OpenMAIC (Open Multi-Agent Interactive Classroom). It runs a phased, confirmation-heavy SOP that routes you into one of three paths: the officially hosted Live Demo (with an sk- access code), a local self-hosted deployment (clone, configure provider keys, start the server), or secondary development on the @openmaic/* SDK. Classroom generation is driven by OpenMAIC's server-side provider configuration, and the skill explicitly asks you to edit local config files yourself rather than writing API keys on your behalf. It suits users who want to generate classrooms directly from messaging apps like Feishu, Slack, or Telegram.
Checks skill config for an accessCode and auto-selects Live Demo mode; routes to the extend branch if your intent is secondary development. Local mode: guides cloning or reusing an existing repo, then choosing a startup mode (pnpm dev / build+start / Docker). Walks you through editing server-side provider config files yourself for the core LLM key plus optional web search, image, video, and TTS features. Verifies the running service with GET /api/health after startup. Submits an async classroom generation job and polls until it succeeds or fails, then returns the classroom link. Extend mode: loads references/extend.md to guide forking/customizing or consuming the @openmaic/* SDK.
- Zero-setup exploration: store an access code from open.maic.chat in config and generate classrooms instantly via the hosted Live Demo
- Self-hosting developers: clone the repo, configure at least one LLM key, and run with pnpm or Docker
- Developers extending OpenMAIC: fork and customize providers, storage, themes, and routes, or build new things with the @openmaic/* SDK
- Offline/intranet training: export classrooms as .maic.zip or Resource Packs with external assets inlined for air-gapped playback
- Users unwilling to edit config files themselves who expect the assistant to fill in API keys — the skill explicitly refuses to write keys on your behalf
- Users without Node.js 20+ and pnpm 10+ who don't intend to use the Live Demo
- Users who configure no LLM provider key at all — generation requires at least one (or an access code in Live Demo mode)
How do you install this skill?
- This is a static review; no commands or generation flows were executed. Conclusions are based on source reading only.
- Live Demo mode sends the sk-prefixed access code as a Bearer header to open.maic.chat; confirm you trust this hosted service before use.
- Several optional features (image/video generation, most web-search providers) rely on overseas services that may be unreachable from mainland China; OpenMAIC supports domestic providers (GLM, MiniMax, Qwen, Kimi, etc.) which are recommended first.
- The skill folder has no independent version or changelog; SDK 0.x versions must be pinned exactly — verify latest registry versions before installing.
- The NEXT_PUBLIC_ persistence token is compiled into public JS and extractable by any visitor; that mode is for localhost/trusted networks only, not public multi-user deployments.
- Shell / CLI
- Network access
- Local filesystem
Node.js >= 20pnpm >= 10OpenMAIC repoat least one LLM provider API key
The source documents OpenClaw/ClawHub installation:
OpenClaw
clawhub install openmaicOr copy manually:
mkdir -p ~/.openclaw/skills
cp -R /path/to/OpenMAIC/skills/openmaic ~/.openclaw/skills/openmaicHow do you use this skill?
Once installed, send your agent any of these to trigger it:
- Install the OpenMAIC skill and teach me quantum physics
- I got an access code from open.maic.chat — configure it and generate an intro-to-Python lesson via the Live Demo
- Help me self-host OpenMAIC: clone the repo, set up an OpenAI key, and start it with Docker
- I want to build on the @openmaic/* SDK — show me how to customize themes and routes
Trigger it by stating your intent to your assistant. The flow is a phased, confirmation-heavy SOP:
- The skill reads accessCode / repoDir / url from skills.entries.openmaic.config in ~/.openclaw/openclaw. as defaults;
- Phase 0 has you pick Live Demo (recommended, code-only, no local setup), local run, or secondary development;
- Local mode proceeds through clone → startup mode → provider keys → health check → generate classroom; extend mode loads extend.md and skips the deployment phases;
- Every state-changing action requires confirmation, but once you clearly ask to generate a classroom, no second confirmation is requested.
What are this skill's strengths and limitations?
- Three usage paths cover the full gradient from instant demo to self-hosting to SDK development
- Confirmation-first SOP keeps every step transparent with no black-box automation
- Sound key-handling design: guides you to edit local configs yourself, never asking you to paste keys into chat
- Rich optional features: web search, image/video generation, and TTS can be enabled per-need
- Local mode has heavy prerequisites: Node.js 20+, pnpm 10+, a cloned repo, and at least one LLM key
- Classroom generation is a long-running async job requiring polling; you may need to check back later
- Depends on OpenClaw's skill mechanism and config file layout; other platforms need adaptation
- No evidence of a dedicated automated test suite for the skill itself
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| OpenMAIC Multi-Agent Interactive Classroom Skill this page | 65 · Recommended | ★ 41k | 11d ago | MIT |
| Add LLM Provider (DeepChat Dev Skill) | 56 · Use with care | ★ 6.4k | 13d ago | Apache-2.0 |
| memU | 50 · Use with care | ★ 15k | 12d ago | NOASSERTION |
| Model Catalog Refresh Skill (agent-service-toolkit) | 60 · Recommended | ★ 4.5k | 21d ago | MIT |
| MAIC Importer Development Skill | 52 · Use with care | ★ 41k | 11d ago | MIT |
Compared to manually deploying OpenMAIC from the repo README, this skill compresses install, config, startup, and generation into a guided, confirmed, phased flow. Compared to a bare self-hosted OpenMAIC instance, it adds the OpenClaw chat entry point (Feishu/Slack/Telegram and 20+ apps) and a hosted Live Demo mode. The source also points to MAIC-UI as a sibling project with richer functionality aimed at high-quality educational UI production.
How did FollowSkills review this skill?
The skill mandates explicit confirmation before every state-changing action, forbids asking for or writing API keys into chat, guides users to edit local config files themselves, and reads access codes only from local config; no covert data flows. Deductions: Live Demo mode transmits the access code as a Bearer header to external open.maic.chat (functionally necessary but still external transmission), rollback paths are not systematically documented, so not full marks.
Phased SOP with clear structure; generate-flow.md contains full polling rules, error-surfacing requirements, capability feature detection, and explicit 'do not work around config errors via request overrides' failure-feedback rules, plus strict output format. Deductions: static review executed nothing; repo CI and unit tests cover product code, not the skill's own key paths — capped at 10 per static calibration.
Three modes (Live Demo / local / extend) with clear boundaries and trigger conditions, including accessCode auto-routing with the extend exception; bilingual with language field; core services open.maic.chat and file.maic.chat are reachable from mainland China and local mode works fully offline of cloud services. Deductions: optional features depend on overseas providers (Google/OpenAI etc.); trigger precision is described textually with no empirical evidence.
Progressive disclosure via SKILL.md plus eight layered references; repo has MIT LICENSE, SECURITY.md, detailed CHANGELOG, release history, and clear maintenance/update paths (publish CI, version-bump checks). Deductions: the skill folder itself has no independent changelog/version, and SDK versions are marked 'current as of this writing' and may drift.
Generation flow, polling, output URL format, quota and error tables are explicit and directly usable; clear marginal value over manual setup. Deductions: static review cannot verify actual correctness/usability of generated classrooms; capped at 7, scored 6.
Most claims trace to repo files (entry points given as file+symbol, real CI workflow with unit and smoke tests), with good fact/inference separation. Deductions: no executed reproduction, repo tests cover the product rather than the skill's key paths; capped at 5, scored 4.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →