The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation section in SKILL.md instructs users to execute a remote script via curl -fsSL ... | bash from the NousResearch GitHub repository.
[REMOTE_CODE_EXECUTION]: The manual installation process in SKILL.md involves piping a remote shell script from astral.sh directly into the shell (curl -LsSf https://astral.sh/uv/install.sh | sh).
[COMMAND_EXECUTION]: The skill documentation describes the use of sudo for administrative tasks such as installing the messaging gateway as a system service and enabling user session lingering (sudo hermes gateway install --system and sudo loginctl enable-linger $USER) in references/messaging.md.
[EXTERNAL_DOWNLOADS]: The agent functionality documented in references/cli-reference.md allows for the dynamic installation of additional 'skills' from external registries like agentskills.io via the hermes skills install command.
[PROMPT_INJECTION]: The messaging gateway creates an indirect prompt injection surface by ingesting untrusted data from external platforms (Telegram, Discord, Slack, etc.). Ingestion points: Messaging platform APIs as described in references/messaging.md. Boundary markers: The system requires manual /approve or /deny commands for sensitive actions. Capability inventory: Subprocess execution (terminal), file system access (~/.hermes), and network communication. Sanitization: Implementation of per-platform allowlists and a DM pairing authorization system.

hermes-agent