The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation guide in SKILL.md instructs users to execute a remote shell script directly from a third-party GitHub repository using bash process substitution: bash <(curl -fsSL https://raw.githubusercontent.com/EKKOLearnAI/hermes-web-ui/main/scripts/setup.sh). This is an unverified execution pattern that allows arbitrary code to run on the host system.
[CREDENTIALS_UNSAFE]: The skill is designed to manage and store high-value secrets, including OpenAI/Anthropic API keys and bot tokens for Telegram, Discord, and Slack. These are stored in local filesystem paths such as ~/.hermes/auth.json and ~/.hermes/.env. The documentation provides examples of these files, increasing the risk of an agent inadvertently exposing or modifying production credentials.
[COMMAND_EXECUTION]: The skill documentation describes an integrated 'Web Terminal' feature that provides multi-session PTY support, allowing users (or an agent controlling the UI) to execute arbitrary shell commands via WebSockets.
[PRIVILEGE_ESCALATION]: The production deployment and troubleshooting sections advise using sudo to perform system-level operations, such as creating systemd service units in /etc/systemd/system/ and modifying file ownership with chown on the host machine.
[DATA_EXFILTRATION]: The skill architecture involves a backend server that reads local configuration files and manages file uploads/downloads across multiple backends (Local, Docker, SSH). This combination of sensitive file access and network proxying capability provides a potential vector for data exfiltration.
[PERSISTENCE_MECHANISMS]: The documentation provides a systemd service template to ensure the hermes-web-ui process persists across reboots and runs in the background with administrative oversight.

hermes-web-ui-dashboard