Hermes Skills System

Summary

The Hermes Skills system is the modular framework for discovering, installing, managing, and distributing agent capabilities. It provides both CLI (hermes skills) and in-chat (/skills) interfaces, supports multiple registries (official, community, self-hosted), and enforces a multi-tiered security/trust model with automated scanning.

Details

Profile Lifecycle

Skills are grouped into “profiles” managed via hermes skills:

• Create: hermes skills create <name> (supports --clone and --no-alias)
• Select: hermes skills use <name> sets a sticky default across sessions
• Inspect: list for overview, show <name> for details
• Portability: export / import package profiles as .tar.gz archives; snapshot export setup.json replicates environments across instances
• Aliasing: hermes skills alias <name> creates wrapper scripts for standalone invocation

Registries and Sources

The Hub aggregates skills from several source types:

Source	Description
Official	Optional skills in the Hermes repo — implicit trust
skills.sh	Vercel’s public directory — community skills, searchable
Well-known	Decentralized: /.well-known/skills/index.json on arbitrary domains
GitHub/Taps	Direct install from repos (e.g., openai/skills/k8s) or custom taps
Marketplaces	ClawHub, LobeHub (converts agent entries), Claude-style marketplaces

Command Interface

• hermes skills browse / search <term>: Discover new capabilities
• hermes skills inspect <slug>: Pre-install details
• hermes skills install <slug>: Fetch and configure
• hermes skills check / update: Identify and apply upstream updates

Security and Trust Model

Every skill undergoes automated scanning for data exfiltration, prompt injection, destructive commands, and supply-chain threats. Four trust levels:

Level	Source	Policy
builtin	Core Hermes code	Always trusted
official	optional-skills/ in repo	Implicit trust, no warnings
trusted	Verified partners (OpenAI, Anthropic)	Permissive execution
community	skills.sh, GitHub, well-known	Requires --force for non-dangerous policy blocks

The --force flag only bypasses “warning” level blocks. A dangerous verdict prevents installation entirely — no override available.

Related

• Hermes CLI
• Hermes Agent Framework
• Hermes Agent SOUL.md
• Hermes Agent Security Model