Skills
skills/rajivpant/synthesis-skills/synthesis-meeting-transcripts/Gen Agent Trust Hub

synthesis-meeting-transcripts

Fail

Audited by Gen Agent Trust Hub on Apr 22, 2026

Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The script optional-workspace-mcp/install-autostart.sh modifies system configuration files to establish background persistence for the workspace-mcp server.
  • Evidence: Writing a LaunchAgent plist to ~/Library/LaunchAgents/com.rajivpant.workspace-mcp.plist on macOS.
  • Evidence: Creating and enabling a systemd user unit at ~/.config/systemd/user/workspace-mcp.service on Linux.
  • Evidence: The skill instructs the user to modify shell configuration files (.zshrc/.bashrc) to add persistent aliases.
  • [REMOTE_CODE_EXECUTION]: The optional-workspace-mcp/start.sh script executes code downloaded from an external public registry at runtime.
  • Evidence: Uses uvx workspace-mcp to download and run the third-party workspace-mcp package from PyPI.
  • [EXTERNAL_DOWNLOADS]: The skill depends on and executes code from a remote source not listed among the trusted organizations.
  • Evidence: Fetches the workspace-mcp package from the PyPI registry.
  • [DATA_EXFILTRATION]: The skill and its associated scripts are designed to perform broad searches and extraction of sensitive meeting data across Google Drive and Gmail.
  • Evidence: fetch-meeting.py uses MCP tools search_drive_files and get_drive_file_content to retrieve full meeting transcripts, including speaker attributions and Gemini-generated notes, which may contain sensitive business or personal information.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Apr 22, 2026, 02:04 PM