TG Canvas Mini App

---

name: tg-canvas description: "Telegram Mini App Canvas with terminal and optional Control UI proxy. Renders agent-generated content (HTML, markdown, A2UI) in a Telegram Mini App; push with tg-canvas push. Includes a JWT-gated browser terminal (server-side PTY/bash — high privilege, shell access to server). Optionally proxies OpenClaw Control UI via ENABLE_OPENCLAW_PROXY=true (off by default; no local files read)." homepage: https://github.com/clvv/openclaw-tg-canvas kind: server metadata: { "openclaw": { "emoji": "🖼️", "kind": "server", "requires": { "bins": ["node", "cloudflared"], "env": ["BOT_TOKEN", "ALLOWED_USER_IDS", "JWT_SECRET", "MINIAPP_URL", "PUSH_TOKEN"] }, "install": [ { "id": "npm", "kind": "npm", "label": "Install dependencies (npm install)" } ] } }

Security FAQ (answers for skill reviewers)

Q: Does the server auto-load ~/.openclaw/openclaw.json or any local credential file? No. The server reads no local files for credentials. OPENCLAW_GATEWAY_TOKEN must be supplied explicitly via environment variable. The relevant code is the ENABLE_OPENCLAW_PROXY config block in server.js — verify it yourself.

Q: What is the default for ENABLE_OPENCLAW_PROXY? Off. The code is const ENABLE_OPENCLAW_PROXY = process.env.ENABLE_OPENCLAW_PROXY === "true"; — it is only enabled if the string "true" is explicitly set in the environment. Omitting the variable leaves it disabled.

Q: What are the terminal/PTY endpoints and how are they authenticated?

• Endpoint: GET /ws/terminal (WebSocket upgrade)
• Auth: JWT verified by verifyJwt() in the upgrade handler — same token issued by POST /auth after Telegram initData HMAC-SHA256 verification against BOT_TOKEN, restricted to ALLOWED_USER_IDS
• If the JWT is missing or invalid the connection is rejected with 401 Unauthorized before a PTY is spawned
• PTY is killed immediately when the WebSocket closes

---

This is a server skill. It includes a Node.js HTTP/WebSocket server (server.js), a CLI (bin/tg-canvas.js), and a Telegram Mini App frontend (miniapp/). It is not instruction-only.

Telegram Mini App Canvas renders agent-generated HTML or markdown inside a Telegram Mini App, with access limited to approved user IDs and authenticated via Telegram initData verification. It exposes a local push endpoint and a CLI command so agents can update the live canvas without manual UI steps.

Prerequisites

• Node.js 18+ (tested with Node 18/20/22)
• cloudflared for HTTPS tunnel (required by Telegram Mini Apps)
• Telegram bot token

Setup

1. Configure environment variables (see Configuration below) in your shell or a .env file.
2. Run the bot setup script to configure the menu button:

   BOT_TOKEN=... MINIAPP_URL=https://xxxx.trycloudflare.com node scripts/setup-bot.js
   

3. Start the server:

   node server.js
   

4. Start a Cloudflare tunnel to expose the Mini App over HTTPS:

   cloudflared tunnel --url http://localhost:3721
   

Pushing Content from the Agent

• CLI:

  tg-canvas push --html "<h1>Hello</h1>"
  tg-canvas push --markdown "# Hello"
  tg-canvas push --a2ui @./a2ui.json
  

• HTTP API:

  curl -X POST http://127.0.0.1:3721/push \
    -H 'Content-Type: application/json' \
    -d '{"html":"<h1>Hello</h1>"}'
  

Security

What the Cloudflare tunnel exposes publicly:

⚠️ Cloudflared loopback bypass: cloudflared (and other local tunnels) forward remote requests by making outbound TCP connections to localhost. This means all requests arriving via the tunnel appear to originate from 127.0.0.1 at the socket level — completely defeating the loopback-only IP check. PUSH_TOKEN is therefore required and is enforced at startup. The loopback check is retained as an additional layer but must not be relied on as the sole protection.

Recommendations:

• Set PUSH_TOKEN — the server will refuse to start without it. Generate one with: openssl rand -hex 32
• Use a strong random JWT_SECRET (32+ bytes).
• Keep BOT_TOKEN, JWT_SECRET, and PUSH_TOKEN secret; rotate if compromised.
• The Cloudflare tunnel exposes the Mini App publicly — the ALLOWED_USER_IDS check in /auth is the primary access control gate for the canvas.
• ENABLE_OPENCLAW_PROXY is off by default. Only enable it if you need Control UI access through the Mini App and understand the implications (see below).

OpenClaw Control UI proxy (optional)

The server can optionally proxy /oc/* to a local OpenClaw gateway, enabling you to access the OpenClaw Control UI through the Mini App.

This feature is disabled by default. To enable:

ENABLE_OPENCLAW_PROXY=true

When enabled, the server:

• Proxies /oc/* HTTP and WebSocket requests to the local OpenClaw gateway.
• If OPENCLAW_GATEWAY_TOKEN is set, injects it as Authorization: Bearer on proxied requests.

The server does not read any local files for credentials — OPENCLAW_GATEWAY_TOKEN must be supplied explicitly via environment variable if needed.

When using /oc/* over a public origin, add that origin to OpenClaw gateway config:

{
  "gateway": {
    "controlUi": {
      "allowedOrigins": ["https://your-canvas-url.example.com"]
    }
  }
}

Terminal (high-privilege feature)

The Mini App includes an interactive terminal backed by a server-side PTY.

⚠️ This grants shell access to the machine running the server, as the process user. Anyone in ALLOWED_USER_IDS can open a bash session and run arbitrary commands. Only add users you trust with shell access to ALLOWED_USER_IDS.

How it works:

• Authenticated users see a Terminal button in the Mini App topbar.
• Tapping it opens xterm.js connected to /ws/terminal (JWT required).
• A PTY (bash) is spawned per WebSocket connection; killed when the connection closes.
• Mobile toolbar provides Ctrl/Alt sticky modifiers, Esc, Tab, arrow keys.

Runtime scope: node-pty spawns a bash process as the server process user. No additional env vars control this; auth is the only gate.

Commands

• tg-canvas push — push HTML/markdown/text/A2UI
• tg-canvas clear — clear the canvas
• tg-canvas health — check server health

Configuration