gemini-api-cli
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Remote Script Execution: The documentation includes instructions to install the CLI tool by piping a script from a vendor-controlled GitHub repository directly into the shell (
curl ... | bash). While this is a standard installation pattern for developer utilities, it involves executing remote code, and it is a recommended practice to review such scripts before execution in your environment. - Indirect Prompt Injection Surface: The skill is designed to ingest and process external content, such as YouTube videos and uploaded documents, which introduces a surface where instructions embedded in that data could potentially influence the agent's behavior.
- Ingestion points: Untrusted data enters the agent context through the
-i(input),-f(file), and--stdinflags (SKILL.md). - Capability inventory: The tool has capabilities to write files to the local system and execute code within managed remote sandboxes (
antigravity-preview-09-2026) (SKILL.md). - Boundary markers: The documentation does not define specific delimiters to isolate external content from system instructions.
- Sanitization: The documentation highlights the use of remote sandboxes and structured JSON output schemas as methods to manage the results of analysis tasks.
Audit Metadata