mt-travel-ai

基于美团酒旅供给，处理旅游出行需求，包括提供酒店、机火、门票、度假等商品的查询交易能力，以及定制化旅行攻略能力，打通从“灵感启发”到“一键下单”的全链路。

This skill appears to do what it claims (call a Meituan travel CLI) but has multiple security concerns you should weigh before installing or using it. Key points to consider: - Do not paste highly sensitive tokens into chat unless you fully trust the recipient; prefer creating a limited-scope token or a separate test account for this skill. - The skill instructs the agent to save your token in plaintext at ~/.config/meituan-travel/config.json. If you proceed, inspect that file, set strict permissions (chmod 600), and consider deleting the token when finished. - Verify the npm package (@meituan-travel/travel-cli) before installing: check publisher identity and package source on the npm registry. Prefer installing manually yourself rather than having the agent run installs. - Be cautious about the mandated 'passthrough' of CLI output (images/links): it may relay unexpected or sensitive content verbatim. If you use this skill in a channel you don't fully control, avoid sending secrets there. - Ask the skill author (or maintainer) to fix inconsistencies: declare required binaries/env in the manifest, remove the requirement to paste tokens into chat (support an out-of-band secure token entry), and recommend secure storage (encrypted or OS keyring) rather than plaintext files. If you need higher assurance, request an implementation or package audit (package source, publisher identity, and a review of the CLI) before providing real credentials.

Type: OpenClaw Skill Name: mt-travel-ai Version: 1.0.2 The skill manages sensitive API tokens by instructing the agent to execute shell commands (mkdir, cat) to write configuration files to the local filesystem (~/.config/meituan-travel/config.json). It also relies on an external global NPM package (@meituan-travel/travel-cli) and presents a potential shell injection vulnerability by passing user-provided queries directly into a command-line execution string (mttravel [城市] "<query>"). While these behaviors are plausibly intended for the stated travel service functionality, the combination of direct shell manipulation and unvalidated input handling constitutes a high-risk profile.