Google Antigravity 模型全自动运维方案。提供多账号自动轮换、优先级调度、会话热更新以及赛博朋克风仪表盘。使用场景包括：(1) 自动化管理多个 Antigravity 账号，(2) 监控配额并自动切换，(3) 在不重启会话的情况下更新模型。

This skill appears to implement the advertised Antigravity account rotation features, but it performs sensitive operations: it reads and updates your OpenClaw auth-profiles (refresh/access tokens), calls external endpoints (a quota API and Google OAuth token endpoint) and executes shell commands (openclaw, curl). Before installing or running: 1) Back up ~/.openclaw/agents/.../auth-profiles.json and any other OpenClaw workspace files it will modify. 2) Inspect the remainder of scripts/rotator.js (performRotation and warmup) to confirm there is no unexpected network exfiltration or unsafe operations — a truncated portion of rotator.js in the package makes a full audit necessary. 3) Consider replacing the bundled clientId/clientSecret with your own OAuth credentials if you want control over which project is used. 4) Run it in an isolated environment or test account first so token updates can't affect production accounts. 5) Note it uses execSync + curl; ensure PATH and config.json are correct to avoid accidental command injection through manipulated config. If you want, I can (a) scan the remaining truncated code for suspicious calls, or (b) produce a short checklist of the exact lines to inspect for exfil/privilege escalation.

Type: OpenClaw Skill Name: antigravity-rotator Version: 1.1.1 The skill is classified as suspicious due to its use of `child_process.execSync` for critical operations and the dashboard listening on `0.0.0.0`. While the `execSync` calls are plausibly needed for the stated purpose of managing OpenClaw models and interacting with Google APIs (e.g., in `scripts/rotator.js` and `scripts/dashboard.js`), this capability inherently carries high risk if inputs were compromised. The dashboard, implemented in `scripts/dashboard.js`, binds to `0.0.0.0` (all network interfaces) instead of `localhost`, potentially exposing its API (which can modify `config.json` and trigger rotations) to other devices on the local network. There is no clear evidence of intentional malicious behavior like data exfiltration or unauthorized backdoors, and the `SKILL.md` does not contain prompt injection attempts with harmful objectives.