agy-staffer
Fail
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill contains explicit instructions to run the companion script unsandboxed and to "request escalated permissions" from the host environment. This is presented as a requirement to access local ports and credential files.
- [CREDENTIALS_UNSAFE]: The instructions specify that the tool requires access to an "OAuth token file" which is typically hidden by the environment's sandbox. Requesting access to these credentials outside of a protected environment poses a risk of exposure.
- [COMMAND_EXECUTION]: The skill executes a Node.js companion script (
agy-companion.mjs) which acts as an interface for the Google Antigravity (agy) CLI tool. - [INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided task text and passes it verbatim to the underlying CLI tool via the
--promptflag. This creates a surface where untrusted user input could influence the behavior of the tool if not properly handled by the external CLI.
Recommendations
- AI detected serious security threats
Audit Metadata