The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches installation scripts and binaries from Nous Research's GitHub repository and Astral's official domain. Evidence: curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh and curl -LsSf https://astral.sh/uv/install.sh in SKILL.md.
[REMOTE_CODE_EXECUTION]: Executes downloaded installation scripts directly via shell pipes to automate environment setup. Evidence: Piped execution patterns (| bash and | sh) are used for automated installation of the agent and the uv package manager.
[PERSISTENCE]: Configures the agent to run as a persistent background service and ensures the CLI is available across sessions. Evidence: Modifies ~/.bashrc to update the system PATH and provides instructions for installing systemd services (Linux) and launchd agents (macOS).
[PRIVILEGE_ESCALATION]: Instructs the user to use sudo for system-level service installation and management. Evidence: Commands like sudo hermes gateway install --system and sudo loginctl enable-linger $USER are used for production deployment.
[COMMAND_EXECUTION]: Provides a --yolo flag that allows the agent to execute terminal commands without manual user approval, increasing the risk of autonomous malicious activity if the agent is compromised. Evidence: The --yolo flag is documented in references/cli-reference.md.
[INDIRECT_PROMPT_INJECTION]: The skill enables a messaging gateway that processes untrusted inputs from third-party platforms while maintaining powerful capabilities. 1. Ingestion points: Receives commands and data from platforms including Telegram, Discord, Slack, WhatsApp, and others as detailed in references/messaging.md. 2. Boundary markers: Implements access control via allowlists and DM pairing codes, but does not specify prompt-level delimiters to prevent instruction injection within user messages. 3. Capability inventory: The agent has full access to the terminal, local file system, network requests, and subagent delegation. 4. Sanitization: Provides optional PII redaction but lacks explicit sanitization or filtering for potential prompt injection attacks embedded in incoming messages.

hermes-agent