The Agent Skills Directory

[COMMAND_EXECUTION]: The dashboard includes core features for arbitrary shell command execution through a browser-based terminal (using node-pty and xterm.js) and a cron job management interface. These features provide a direct path for executing arbitrary code on the host system.
[PRIVILEGE_ESCALATION]: The installation instructions require sudo permissions to install system dependencies. More critically, the recommended systemd service configuration (hermes-control.service) runs the application process with root privileges (User=root).
[EXTERNAL_DOWNLOADS]: The setup process involves cloning the entire application source code from an external, unverified GitHub repository (https://github.com/xaspx/hermes-control-interface.git) and installing multiple third-party Node.js dependencies.
[DATA_EXFILTRATION]: The skill facilitates access to sensitive files and credentials. The file explorer provides read/write access to specified directories, and the configuration management system handles various third-party API keys (e.g., OPENROUTER_API_KEY, TELEGRAM_BOT_TOKEN, TAVILY_API_KEY). A compromise of this interface would lead to complete data exposure.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes untrusted data.
Ingestion points: External chat messages via /api/chat/send and local file content via /api/files/read.
Boundary markers: None explicitly described in the instruction set.
Capability inventory: Significant capabilities including arbitrary terminal execution, file system modification, and cron job management.
Sanitization: Mentions basic path validation and scoping for the file explorer, but general sanitization for message processing is not detailed.

hermes-control-interface-dashboard