hermes-desktop-companion

Fail

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides instructions to grant the user passwordless sudo access by writing to /etc/sudoers.d/hermes-install, which significantly compromises system security by allowing processes to run administrative commands without authentication.
  • [EXTERNAL_DOWNLOADS]: The skill directs users to download and execute platform-specific binaries from an external GitHub repository (fathah/hermes-desktop).
  • [REMOTE_CODE_EXECUTION]: The first-run setup process involves downloading and executing a remote installation script from the NousResearch repository.
  • [COMMAND_EXECUTION]: The skill implements slash commands (/shell and /code) that provide the AI agent with capabilities to execute arbitrary terminal commands and code on the host system.
  • [COMMAND_EXECUTION]: The skill instructs users on how to bypass operating system security controls, specifically macOS Gatekeeper quarantine attributes (xattr -cr) and Windows SmartScreen warnings for unsigned binaries.
  • [DATA_EXFILTRATION]: The combination of arbitrary shell access and the ingestion of untrusted web data creates a significant surface for indirect prompt injection and the exfiltration of sensitive local data like environment variables or configuration files.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
May 17, 2026, 04:22 PM
Security Audit — agent-trust-hub — hermes-desktop-companion