The Agent Skills Directory

[DATA_EXFILTRATION]: The bridge service exposes sensitive user data through unauthenticated API endpoints. While the message-sending API is protected by a secret, the GET /api/contacts and GET /api/rooms endpoints in scripts/wechat-bridge.js lack authentication checks, allowing unauthorized access to the user's full WeChat contact list, contact IDs, and group names.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it forwards external content from WeChat to the AI agent without proper sanitization or boundary markers.
Ingestion points: WeChat message text is captured in scripts/wechat-bridge.js via message.text().
Boundary markers: Absent. Messages are forwarded as raw text within a JSON payload to the OpenClaw gateway.
Capability inventory: The service has the capability to read local files (via FileBox.fromFile) and send them to WeChat, as well as fetch and send remote images (via FileBox.fromUrl).
Sanitization: Only @mention strings are removed from the text; no further escaping or instruction-filtering is performed before the content reaches the agent context.
[EXTERNAL_DOWNLOADS]: The skill utilizes the PadLocal protocol for WeChat integration. It downloads and uses the wechaty-puppet-padlocal package and communicates with pad-local.com to maintain the bridge connection.

wechat-channel