xui-hub v1.0.0 发布 🚀 现在可以让 AI 帮你管 3x-ui 了
🤖 自带 MCP 服务:Claude、Cursor 等 AI 客户端直接查客户、建客户、续期、开线路(支持 OAuth 和 API 令牌)
🔑 开放 API:发卡站或机器人带令牌就能自动开通
📊 流量统计:按客户 / 按面板看 1–90 天用量和每日走势
☁️ 异地备份:每日自动传到 Cloudflare R2 等 S3 兼容存储,后台一键恢复
⬆️ 自动发现新版本,后台直接点更新
🌐 中文 / English 双语界面
另外还有:设备数上限、疑似分享提醒、每月自动重置流量、一键重启面板和 Xray 内核。
Manage multiple 3x-ui panels from one hub, now with MCP so your AI agent can run it.
MIT 开源,一行脚本或 Docker 部署 👇
https://t.co/jTbctCuzhi
#3xui #xray #MCP #selfhosted
@cursor_ai Same trust issue shows up in agent skills: people install by name. We scanned 66 popular ones (57 shown: 24 safe, 17 low, 16 medium). Rules flagged 9 as high/critical; 7 were false positives after reading the source. The other 2 are medium. Every flag shows its code line.
@Slonski_rt +1 on reading SKILL.md. We scanned 66 popular skills (57 shown: 24 safe, 17 low, 16 medium). Rules flagged 9 as high/critical; 7 were false positives (a "never read ~/.ssh" blocklist matched as credential theft). The other 2 are medium. Every flag shows its code line.
@mirku21 Before installing skills or hooks: we scanned 66 popular ones (57 shown: 24 safe, 17 low, 16 medium). Rules flagged 9 as high/critical; 7 were false positives (a "never read ~/.ssh" blocklist matched as credential theft). The other 2 are medium. Every flag shows its code line.
Is the AI skill you installed safe?
A GitHub issue reports Postman's desktop app put plugins and a self-approved startup hook into Claude Code and Codex unasked. No exfiltration shown.
We scanned 66 popular skills; each finding links its code line:
https://t.co/IVu2iGb9BO
We scanned 66 popular AI coding agent skills. Rules flagged 9 as high/critical: 7 were false positives once we read the source (a "never read ~/.ssh" blocklist matched as credential theft), the other 2 are medium. Every flag shows its code line: https://t.co/IVu2iGaBMg
@BleepinComputer Expect a wave of scanners for agent tooling now. Lesson from running rules over 57 popular agent skills: a security checklist describing an attack matches as the attack itself. 7 of our 9 high/critical hits turned out to be false positives, so show the code line.