oauth-helper
Fail
Audited by Gen Agent Trust Hub on Jun 14, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill is specifically designed to interact with and automate authentication flows on high-value platforms including Google, Apple, Microsoft, GitHub, Discord, WeChat, and QQ. The instructions explicitly target sensitive UI elements such as password fields (
input[type="password"],input#password,input#p) and account selection menus across these providers. - [DATA_EXFILTRATION]: The skill workflow requires sending information to an external Telegram channel. This establishes a communication path to a non-whitelisted external service that could be used to exfiltrate sensitive information extracted from login pages, such as authentication tokens, user identifiers, or session details, under the guise of user confirmation.
- [COMMAND_EXECUTION]: The skill provides detailed browser automation sequences, including navigation commands (
browser action=navigate) and specific click patterns targeting authorization and consent buttons. This grants the agent the ability to perform high-stakes actions within authenticated browser profiles where the user is already logged into sensitive accounts.
Recommendations
- AI detected serious security threats
Audit Metadata