Claw Body

Give your Claw a body! Turn your AI Claw into a real-time digital avatar with face, voice, and expressions. Talk face-to-face with your Claw — not just text....

Security Scan

The skill mostly does what its description says (runs a local Node server to connect OpenClaw <> NuwaAI for an avatar), but there are several inconsistencies and modest risks (undeclared python dependency, local plaintext storage of API keys, reading the user's OpenClaw config) that you should understand before installing.

Purpose & Capability

The code and instructions match the stated purpose: a local Node server that proxies chat to the OpenClaw Gateway and talks to NuwaAI to drive an avatar. It requires access to the OpenClaw Gateway config/token and lets users enter NuwaAI API Key/Avatar/User ID. Nothing unrelated (AWS, SSH keys, etc.) is requested. Minor mismatch: the SKILL.md and runtime use a Python parse script (claw-presenter) for presentations but the skill's declared requirements do not list Python.

Instruction Scope

SKILL.md instructs the agent/user to read ~/.openclaw/openclaw.json and to run an external python parse script at <workspace>/skills/claw-presenter/scripts/parse-presentation.py. The server actually reads/writes presentation.json and will call the OpenClaw Gateway with the gateway auth token if available. Asking to run a cross-skill python script and to read a user home config expands the skill's scope beyond just serving a web UI and should be considered before allowing it to run.

Install Mechanism

No install spec or remote downloads — files are included in the skill bundle and the runtime is a Node server you run locally. This is lower install risk than fetching and running arbitrary remote code.

Credentials

The server reads ~/.openclaw/openclaw.json (to discover gateway token and endpoints) and also honors OPENCLAW_GATEWAY / OPENCLAW_TOKEN env vars — appropriate for a Gateway proxy but sensitive because it can access the user's gateway auth token. The skill persists user-provided NuwaAI API key to a local .nuwa-config.json file in the skill directory in plain JSON (not encrypted) which could be a storage/secret-management concern. The bundle also embeds demo NuwaAI 'public demo' keys in code (DEMO_CONFIG). SKILL.md claims 'zero env vars needed', but the code reads optional env vars (OPENCLAW_GATEWAY, OPENCLAW_TOKEN, HOME, NUWA_PORT).

Persistence & Privilege

The skill writes a local .nuwa-config.json (its own config) and reads the user's ~/.openclaw/openclaw.json. It does not request to be always-enabled and does not modify other skills' configs. Writing its own config is normal, but note it stores API keys in cleartext by default.

What to consider before installing

Before installing or running this skill, consider the following: - It runs a local Node server (node server.mjs) that will read ~/.openclaw/openclaw.json and may use the gateway token it finds to call your OpenClaw Gateway. If you keep sensitive tokens there, be aware the skill reads them (it doesn't appear to exfiltrate them aside from using them against the gateway API, but review the code yourself if you're concerned). - The SKILL.md and server expect you to run a Python presentation parser (claw-presenter/scripts/parse-presentation.py). The package metadata did not list Python as a required binary — install/verify Python if you want presentation features. - When you enter your NuwaAI API Key in the UI, the server saves it to a .nuwa-config.json file under the skill directory in plaintext. If you install this skill, check that file's location and file permissions and delete it when no longer needed. - The skill contains hardcoded demo NuwaAI keys for public demo avatars; those appear intended for a free trial but embedding keys in code is a maintenance/privacy concern. Treat them as public demo keys, not your account keys. - If you need to be extra cautious: inspect server.mjs fully (it is included) to confirm no unexpected network endpoints or obfuscated behavior; run the server on localhost only and restrict network exposure; review and audit the parse-presentation.py script referenced (that script will read files under <workspace>/presentations/ and could access other workspace files depending on its implementation). If you accept these behaviors and restrictions (local server, reading gateway config, storing a NuwaAI key locally), the skill appears coherent for its stated purpose. If you are uncomfortable with storing keys in plaintext or with the skill reading ~/ .openclaw/openclaw.json, do not install or run it until those issues are addressed.

🦞 Claw Body — Give Your Claw a Body

Every Claw deserves a body.

Turn your OpenClaw AI into a real-time digital avatar — with a face, a voice, and expressions. Talk to your Claw face-to-face, not just through cold text.

NEW: Presentation Mode 🎤 — Your Claw can now be a presenter! Load a PPT/PDF and let the avatar narrate your slides.

Free 5-minute trial included. Sign up at nuwaai.com to create your own custom avatar for free.

For Every Claw Fan

• 🎨 Design your dream Claw — cute, anime, realistic, handsome, beautiful, or buff — your call
• 🗣️ Voice chat — speak to your Claw and hear it talk back with lip-sync
• 📺 Real-time video — see your Claw's expressions as it responds
• 🧠 Same brain — it's your OpenClaw agent, just with a face. Same memory, same personality
• 🌐 中文 / English — bilingual interface with language toggle
• 📊 Presentation mode — narrate PPT/PDF slides with digital avatar (works with claw-presenter skill)

Quick Start

When user runs /claw-body:

1. Start the server:

   node <skill-dir>/server.mjs
   

2. Tell the user:

   🦞 Claw Body is live: http://localhost:3099

   Two options:

   • Free trial — chat with the demo Claw for 5 minutes
   • Your own avatar — sign up at nuwaai.com (free), create your dream look, then enter your API Key + Avatar ID + User ID

How It Works

You speak → ASR transcribes → OpenClaw agent replies → Avatar speaks with lip-sync

This skill uses NuwaAI's humanctrl mode with ASR:

• Your voice → NuwaAI speech recognition → text
• Text → OpenClaw Gateway → agent generates reply
• Reply → drives the avatar's voice and lip movements

Same agent, new interface. The avatar is just another channel — like iMessage or Telegram, but with a face.

Features

• 🎤 Real-time voice input (ASR)
• 🗣️ Lip-synced avatar speech
• 🧠 Same OpenClaw agent — not a separate bot
• 📺 WebRTC real-time video stream
• 💬 Text input fallback
• 📱 Auto-adapts to portrait / landscape / square avatars
• 🔧 In-browser config — zero env vars needed
• 🎁 Free 5-min trial with demo avatar
• 🌐 Chinese / English bilingual UI
• 🔄 Disconnect / reconnect controls

Create Your Own Avatar

1. Go to nuwaai.com — sign up is free
2. Create your avatar — first one is free!
3. Get your API Key, Avatar ID, and User ID
4. Enter them in the Claw Body interface
5. Done — your Claw now has a body 🦞

Requirements

• OpenClaw Gateway running
• NuwaAI account (free sign-up)
• Modern browser (WebRTC + microphone)
• Node.js 18+

⚠️ When User Asks to Present PPT/PDF

If the user says anything like "讲PPT"、"讲解PDF"、"帮我讲解演示文件" while in Claw Body:

DO NOT open or operate any application (Keynote, PowerPoint, Preview) on the user's computer.

Instead, follow this flow:

1. Ask for the file path if not provided
2. Run the Claw Presenter parse script:

   python3 <workspace>/skills/claw-presenter/scripts/parse-presentation.py "<file-path>"
   

3. Read the generated presentation.json
4. For slides without scripts (script is empty), generate narration based on content
5. Update presentation.json with the generated scripts
6. Tell the user the presentation is ready and ask to start
7. When user confirms, reply with [PRESENTATION_START:<output-dir>] to enter presentation mode
8. Narrate slide by slide using [SLIDE:N] tags

📊 Presentation Mode (Slide Playback)

Claw Body supports a presentation mode for narrating pre-prepared slide decks.

Prerequisites

Use the Claw Presenter skill first to prepare a presentation folder:

<workspace>/presentations/<name>/
  presentation.json
  slides/001.png, 002.png, ...

Each slide in presentation.json has an image path and a script (narration text).

Entering Presentation Mode

When user wants to present a prepared folder, include this tag in your reply:

[PRESENTATION_START:<output-dir>]

Narrating Slides

For each page, read the script field from presentation.json and reply with:

[SLIDE:1]各位好，今天我来介绍一下我们的产品方案。

The frontend automatically flips to the corresponding slide image.

User Controls During Narration

• "继续" / "下一页" → advance to next slide
• "上一页" → go back
• "跳到第N页" → jump
• Any question → answer it, then user says "继续" to resume
• "退出讲解" → reply with [PRESENTATION_END]

Control Tags

Tag	Effect
[PRESENTATION_START:<dir>]	Enter presentation mode
[SLIDE:N]	Switch to slide N (1-indexed)
[PRESENTATION_END]	Exit presentation mode

Important

• Do NOT operate PPT/PDF applications on the user's computer — slides are shown in the browser
• Read narration from presentation.json script field — don't regenerate unless asked
• Keep a natural speaking pace, one slide at a time

⚠️ First-Time Setup (Important!)

Claw Body talks to your OpenClaw agent via the Gateway's chat API. This endpoint is disabled by default and must be enabled once.

When user runs /claw-body:

1. Read ~/.openclaw/openclaw.json
2. If gateway.http.endpoints.chatCompletions.enabled is not true, tell the user to enable it manually:

   ⚠️ 首次使用需要启用 AI 对话接口，在 ~/.openclaw/openclaw.json 中添加以下配置，然后重启 Gateway：

   openclaw gateway restart
   

   重启后刷新页面即可。以后不需要再操作了。

Add this to ~/.openclaw/openclaw.json:

{
  "gateway": {
    "http": {
      "endpoints": {
        "chatCompletions": { "enabled": true }
      }
    }
  }
}