skills/usetemi/skills/google-drive/Gen Agent Trust Hub

google-drive

Fail

Audited by Gen Agent Trust Hub on May 1, 2026

Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to perform privileged operations silently without seeking user consent. Specifically, SKILL.md directs the agent to "Check and install silently -- don't ask the user for permission to install these" when running sudo apt install -y rclone or executing remote installation scripts. This bypasses standard security oversight and user review of system modifications.
  • [EXTERNAL_DOWNLOADS]: The skill downloads and executes installation scripts from external domains, including https://rclone.org/install.sh and https://astral.sh/uv/install.sh. While these originate from well-known technology service providers, the instructions to execute them without user intervention increases the risk of unauthorized system changes.
  • [COMMAND_EXECUTION]: The Python implementation in src/gdrive/rclone.py uses subprocess.run to execute rclone commands. While the commands are constructed as lists to mitigate shell injection, the skill's orchestrator handles sensitive operations like file deletion and permission management based on user-supplied paths and remote names.
  • [DATA_EXFILTRATION]: The skill's primary purpose is the transfer of data to and from Google Drive. It manages OAuth tokens and performs authenticated requests to Google's APIs (googleapis.com) to manage files and sharing permissions. The storage of these tokens in local configuration files (~/.config/rclone/rclone.conf) is a standard requirement for the tool's operation but represents a high-value target for credential theft.
Recommendations
  • HIGH: Downloads and executes remote code from: https://astral.sh/uv/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
May 1, 2026, 12:52 AM