delegate
Pass
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local script
antigravity.pyviauv runto manage agent tasks, passing user-provided input as arguments. - [REMOTE_CODE_EXECUTION]: Facilitates code execution in a managed sandbox environment provided by Google Gemini, which is a trusted service.
- [EXTERNAL_DOWNLOADS]: Allows the user to specify external GitHub repositories to be mounted in the remote sandbox via the
--repoflag. - [PROMPT_INJECTION]: Ingests potentially untrusted output and traces from a remote agent. Ingestion points: Agent output and traces in Phase 4. Boundary markers: None. Capability inventory: Local command execution and remote sandbox execution. Sanitization: None. However, the source is a trusted managed service (Google), mitigating the risk.
Audit Metadata