setup
Warn
Audited by Gen Agent Trust Hub on May 13, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads and installs the 'gbrain' package directly from a GitHub repository ('github:garrytan/gbrain') using the Bun package manager.
- [COMMAND_EXECUTION]: Executes shell scripts to discover markdown repositories by scanning broad system paths including '/data/', '~/git/', and '~/Documents/*'. This can expose the structure of sensitive user directories.
- [COMMAND_EXECUTION]: Configures system-level persistence mechanisms through the 'gbrain autopilot --install' command, which writes launchd plists on macOS, systemd service units on Linux, or crontab entries, allowing code to run automatically in the background.
- [CREDENTIALS_UNSAFE]: Requests and processes a PostgreSQL connection URI that explicitly contains the database password in the clear ('postgresql://postgres.[ref]:[password]@...').
- [DATA_EXFILTRATION]: Contains instructions to scan for and 'import' content from various local directories into a database. While intended for knowledge management, this capability can be used to aggregate sensitive information from the local file system.
Audit Metadata