小红书舆情哨兵

面向中文用户的小红书舆情巡检技能。按关键词抓取搜索结果，判断登录墙/验证码/空结果状态，输出结构化笔记、热度/风险/竞品信号和现场截图。使用前需配置 XHS_COOKIE 与消息接收渠道（如 Telegram、企业微信）。适用于品牌口碑监控、竞品观察、热点追踪与内容研究。

This skill appears to do what it says: run a headless browser, search Xiaohongshu, capture screenshots and extract simple signals. Before installing or running it: (1) confirm you trust the source — the registry metadata omits required env vars but SKILL.md requires XHS_COOKIE and optional CHANNEL/TARGET_ID; (2) avoid storing XHS_COOKIE or other secrets in generic helper scripts or public repos — the script will read run_xhs_monitor.sh in several locations; (3) ensure you have Node, puppeteer-core and a Chrome/Chromium binary available; (4) review the remainder of monitor.js (the file was partially shown) to verify whether it implements any outbound delivery (Telegram/WeCom) and how TARGET_ID is used; (5) if you must protect credentials, run this in an isolated environment or supply cookies only via an env var set for the runtime. If you want, I can scan the rest of monitor.js (the truncated tail) to confirm there are no unexpected network endpoints or data exfiltration paths.

Type: OpenClaw Skill Name: xhs-sentry Version: 1.3.4 The skill exhibits suspicious behavior by searching the user's filesystem (including the home directory and OpenClaw workspace) for credentials stored in other scripts' files to extract the 'XHS_COOKIE' (scripts/monitor.js). It also utilizes prompt injection techniques by appending 'System Hints' to its console output, which are designed to override the AI agent's default behavior and force specific formatting (SKILL.md). While these actions are documented as features for user convenience and output consistency, the automated harvesting of secrets from unrelated files and the use of instruction-injection patterns are high-risk behaviors.