The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the agent using a shell script downloaded from a remote GitHub repository and piped directly into bash (curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash).
[PERSISTENCE]: The framework includes built-in mechanisms for persistence, such as installing itself as a background service (hermes gateway install) and a dedicated cron system for scheduled task execution (hermes cron).
[PRIVILEGE_ESCALATION]: The skill documents the --yolo flag and approvals.mode: off configuration, which allow the agent to skip user approval for dangerous system commands. It also provides instructions for disabling default security redaction of API keys and secrets (security.redact_secrets false).
[INDIRECT_PROMPT_INJECTION]: The agent framework has a large attack surface for indirect injection as it ingests data from numerous untrusted ingestion points (Telegram, Discord, Slack, WhatsApp, Signal, Email, and Webhooks). This is combined with powerful capabilities including shell command execution (terminal tool), file manipulation (file tool), and sandboxed Python execution (code_execution), posing a risk if untrusted input contains malicious instructions. The skill documents redaction features as mitigations, but these can be manually disabled.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and execution of external content, including a hub for installing additional skills via IDs or direct URLs (hermes skills install ID).

hermes-agent