xiaoai-bridge

小米小爱音箱语音指令桥接。截取小爱音箱的语音消息，转换为 AI 助手指令，并通过 TTS 回复。支持触发词过滤、自动去重、后台监听。适用于通过小爱音箱语音控制 OpenClaw 助手、智能家居联动、语音任务执行等场景。

This skill does roughly what it claims (connect to Xiaomi, poll voice messages, output JSON, and play TTS), but there are important red flags you should address before installing: - Do NOT use credentials included in the bundle (scripts/.mi.json). That file contains a passToken/serviceToken and device identifiers; treat it as compromised and remove it. Always use your own MI_PASS_TOKEN or a dedicated account. - The package metadata omitted required env vars — expect to provide MI_USER_ID, MI_PASS_TOKEN (recommended) or MI_PASSWORD, and MI_DEVICE_ID. Confirm these requirements with the code before running. - Inspect the @mi-gpt/miot dependency yourself (package-lock points to a Tencent mirror). If you cannot verify the registry/mirror, consider installing dependencies from the official npm registry or review the package source. - Because the listener continuously polls and prints conversation text, run the skill in an isolated environment (container or VM) to avoid accidental data leakage. Review logs carefully — the script will output user voice text and device info. - Replace any use of a real personal Xiaomi account with a dedicated/test account if possible. If you ever used the included tokens, rotate credentials immediately. - Investigate the prompt-injection warning in SKILL.md (base64 or other embedded payloads) and remove any unexpected or encoded content. If you are not comfortable auditing the dependency (@mi-gpt/miot) and removing the embedded .mi.json first, treat this skill as unsafe to install in a production environment.

Type: OpenClaw Skill Name: xiaoai-bridge Version: 1.0.0 The skill bundle is classified as suspicious due to two critical vulnerabilities. Firstly, the `scripts/.mi.json` file contains hardcoded Xiaomi account credentials (userId and passToken), which is a severe accidental credential leak of the developer's secrets. Secondly, the `SKILL.md` documentation provides example code that uses `child_process.exec` to execute `xiaoai-listen.js speak "${text}"`. If the `text` variable, originating from AI agent responses to voice commands, contains shell metacharacters, it could lead to arbitrary command execution (RCE) due to lack of input sanitization in the recommended integration pattern.