The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation guide includes a command to download and pipe a shell script directly into a root shell (curl -sSL ... | sudo bash) from the repository OnlyTerp/hermes-optimization-guide. This allows arbitrary code execution on the host system without prior user review of the script content.
[PRIVILEGE_ESCALATION]: The skill provides instructions for multiple high-privilege operations, including the use of sudo for package management, user creation (useradd), and systemd service installation in /etc/systemd/system/.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs software from several non-whitelisted sources, including a VPS bootstrap script from a personal GitHub account and various NPM packages such as @nousresearch/hermes-agent and multiple MCP servers.
[PERSISTENCE]: The skill guides users through setting up systemd services to maintain the agent's operation across reboots and includes configuration for cron jobs to execute tasks at scheduled intervals.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a significant attack surface for indirect prompt injection through its integration with external data sources.
Ingestion points: Untrusted data enters the agent context through MCP servers (filesystem, github, brave_search) and via direct CLI pipes when processing external files.
Boundary markers: The configuration examples do not include explicit delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill enables high-risk tools including read_file, write_file, and execute_command alongside various MCP tool integrations.
Sanitization: There is no evidence of input validation, filtering, or sanitization for content retrieved from external search results, code repositories, or local files.

hermes-agent-optimization