Security audit
Security checks for vulnerabilities and agentic risk
This is a single informational Markdown skill about Gemini on Android, with no executable code or hidden install behavior.
Installing this skill should be low risk as documentation, but treat its automation examples as guidance only: do not let any assistant submit forms, add shopping items, use emails, or act across apps without explicit review and confirmation.
The trigger phrases are very broad and generic, such as 'Gemini', '安卓AI', and '手机AI', which can easily overlap with normal user conversation or unrelated requests. This can cause unintended skill activation and routing, especially because the skill covers system-level and cross-application capabilities that may expose sensitive context or cause confusing automation flows.
This section promotes cross-app automation, screenshot interpretation, email reading, and shopping-cart actions without warning that personal content, account state, or purchasing actions may be involved. That omission is risky because users may not realize the skill can operate on sensitive data or produce side effects across apps.
The browser-assistant section advertises automatic browsing and complex form filling without warning about submission of passwords, payment data, personal identifiers, or legal/medical information. In context, this is particularly concerning because browser forms often contain highly sensitive data and users may overtrust 'automatic' completion behavior.
The guidance '直接说"帮我..."' encourages invocation through an extremely ambiguous phrase that is common in everyday speech. In an environment with multiple skills or assistants, this increases the chance of accidental activation and unintended execution of automation-oriented behavior.
No suspicious patterns detected.
