ogt-cli-gemini

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS] (LOW): The skill installs the @google/gemini-cli package via npm and gemini-cli via Homebrew. These are associated with a trusted organization (Google) according to the [TRUST-SCOPE-RULE], though users should still verify the package's integrity at install time.
  • [COMMAND_EXECUTION] (MEDIUM): The skill provides a wrapper for a CLI tool that has built-in capabilities to execute terminal commands. Specifically, the documentation mentions the --yolo flag, which auto-approves tool calls. If the agent processes malicious instructions from an external source (indirect injection), it could execute arbitrary commands without user intervention.
  • [REMOTE_CODE_EXECUTION] (MEDIUM): The combination of 'Web Fetching', 'Google Search grounding', and 'Shell Commands' allows the agent to retrieve remote data and potentially execute it as code. This path is explicitly supported by the CLI's configuration.
  • [UNVERIFIABLE_CODE] (MEDIUM): The skill references a script located at {baseDir}/scripts/run-gemini-task.cjs for sub-agent delegation. This file was not provided in the skill package for security auditing, making its actual behavior unverifiable.
  • [INDIRECT_PROMPT_INJECTION] (LOW):
  • Ingestion points: The tool reads local files, project directories, and GEMINI.md for context.
  • Boundary markers: No specific delimiters or safety warnings for the LLM are provided in the skill instructions.
  • Capability inventory: Includes file-write, shell command execution, and network access.
  • Sanitization: There is no evidence of input sanitization before passing project file content to the Gemini model.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 06:29 PM