The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides a command to download and execute an installation script from https://hermes-agent.nousresearch.com/install.sh by piping the output of curl directly into bash. This pattern allows for arbitrary code execution from a remote server without prior inspection.
[EXTERNAL_DOWNLOADS]: Users are encouraged to clone and install code from a large number of third-party GitHub repositories belonging to various user accounts (e.g., wondelai, tlehman, builderz-labs). These resources represent a supply chain risk as they originate from unverified sources.
[DYNAMIC_EXECUTION]: The documentation highlights tools such as hermes-dojo and hermes-skill-factory which are designed to autonomously generate or modify executable skills at runtime based on observed agent behavior or performance metrics.
[PERSISTENCE]: The skill includes instructions for configuring recurring cron jobs within the agent's profile (config.yaml). This mechanism can be used to maintain background execution of tasks across sessions.
[INDIRECT_PROMPT_INJECTION]: The skill sets up surfaces for processing untrusted data from external sources like Nextcloud, Spotify, and web search results. This data could contain hidden instructions that influence the agent's behavior.
Ingestion points: Data ingested via hermes-nextcloud, wondelai search skills, and workflow observation by hermes-skill-factory.
Boundary markers: None defined in the skill instructions to separate data from commands.
Capability inventory: The skill utilizes execute_code, subprocess.run, and file writing operations.
Sanitization: There are no mentions of input validation or sanitization for data retrieved from external sources.

awesome-hermes-agent-ecosystem