推送消息数据助手（友盟U-Push）

友盟推送后台管理助手（只读查询）。帮助获取应用列表、查询推送数据等。**使用前需用户提供 Cookie**：请访问 https://upush.umeng.com 登录后，在浏览器开发者工具的 Network 标签中复制 Cookie 并提供给系统。Use when working with Umeng push...

This skill implements read-only queries for Umeng Push and expects you to provide your browser session cookie (ctoken and login id). Before installing or running it: 1) Do NOT paste full cookies into any public chat — that shares a live session token. Prefer storing cookies only locally using the provided scripts and run them locally, not in a hosted agent environment. 2) Inspect the bundled Python scripts (especially auto_get_cookie.py, browser_cookie.py, and any network code) yourself — auto_get_cookie can access your browser cookies beyond just Umeng's session. 3) Note the inconsistent file usage (cookie.json vs cookie.txt and different paths); verify where cookies will be saved and ensure file permissions and storage location are acceptable. 4) If possible, create a low-privilege/test account or use ephemeral session cookies instead of your main account. 5) Avoid running the automatic cookie retrieval unless you trust the code and run it in an isolated environment. 6) The skill includes a 'security_interceptor' blocking write APIs (sendMsg, updateApp, etc.), which is good, but that does not eliminate the cookie-exposure risk. If you are not comfortable reviewing code, do not provide session cookies to this skill.

Type: OpenClaw Skill Name: umeng-push-helper Version: 1.0.0 The bundle provides tools for Umeng Push management but contains high-risk vulnerabilities and documentation inconsistencies. Specifically, `api_request.py` includes a `custom` command that allows the agent to make HTTP requests to any arbitrary URL while automatically attaching the user's sensitive Umeng session cookies, as the `make_request` function lacks domain validation. Additionally, while `SKILL.md` claims automatic cookie extraction is no longer supported, the bundle includes `auto_get_cookie.py` and documentation (`CHANGELOG_COOKIE_UPDATE.md`) promoting it as the recommended method. These factors, combined with the broad capability to query sensitive account data, create a significant surface for credential exfiltration if the agent is targeted by prompt injection.