google-agents-cli

Warn

Audited by Gen Agent Trust Hub on May 16, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires installing the google-agents-cli package via pip and uvx. The package name uses a prefix that may be misleading given the skill is authored by a third party (ara.so).
  • [REMOTE_CODE_EXECUTION]: The installation command uvx google-agents-cli setup downloads and executes code from a remote source on the host system. Additionally, the eval run command executes Python code provided in evaluation configuration files.
  • [COMMAND_EXECUTION]: The skill relies on various shell commands through the agents-cli tool for tasks such as infrastructure provisioning, project scaffolding, and deployment to Google Cloud.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface (Category 8) due to its processing of external data and configuration.
  • Ingestion points: evalsets/basic.yaml (test case inputs), gs://my-bucket/docs (RAG source data).
  • Boundary markers: Absent from the documented configuration patterns.
  • Capability inventory: agents-cli eval run (code execution), agents-cli deploy (cloud service management), agents-cli run (agent interaction).
  • Sanitization: No sanitization or validation of external content is mentioned.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 16, 2026, 05:27 PM