Educational selection guide

Hermes Agent resources / extension surfaces

Tools, skills, and plugins: choose an extension path

Hermes Agent’s official documentation distinguishes tools and toolsets, on-demand skills, and plugins that add custom tools, hooks, and integrations. This guide turns those distinctions into a review checklist; it is not a replacement for current Nous Research documentation and does not publish install commands that may change.

Similar words, different responsibilities

Start by identifying the extension surface you are evaluating. The labels below follow the current official docs and intentionally avoid treating every add-on as interchangeable.

01 / Tools

Functions the agent can call

Tools extend an agent’s capabilities. Hermes groups built-in tools into toolsets that can be enabled or disabled per platform, so the useful question is not only what a tool does but when it is enabled.

Official tools and toolsets page ↗

02 / Skills

On-demand knowledge and procedure

Skills are knowledge documents the agent can load when needed. The official skills page describes progressive disclosure and local skill directories; a skill is not automatically proof that its procedure is safe for every project.

Official skills system page ↗

03 / Plugins

Custom tools and integration hooks

Plugins are the documented path for adding custom tools, hooks, commands, and integrations without modifying Hermes core. They can also bring dependencies, data files, and bundled skills into the review.

Official plugins page ↗

Five questions before you enable or install

This original checklist is a decision aid, not a security certification. Answer it with the exact version, source, profile, host, and data involved, then confirm any platform-specific behavior in the official docs.

  1. Permissions: What can the tool, skill, plugin, or its host read, write, execute, send, or control? Scope the enabled toolsets and user authorization to the smallest useful boundary, and note whether a project-local extension is trusted by placement.
  2. Dependencies: Which runtime packages, binaries, services, MCP servers, or provider accounts are required? Record sources and versions, check the project’s preparation notes, and treat optional dependencies as part of the supply-chain review rather than an invisible detail.
  3. Data destinations: Where do prompts, files, tool results, credentials, logs, and generated artifacts go? Distinguish local storage, a terminal backend, an MCP server, an inference provider, and a messaging or browser surface before using real data.
  4. Review: Can you inspect the manifest, source, requested capabilities, configuration, and proposed action before activation? Start with a disposable fixture, keep human approval for consequential changes, and verify what the agent actually did rather than relying on its summary.
  5. Rollback: What is the disable, remove, restore, or profile-isolation path if the extension misbehaves? Save the relevant configuration, preserve a clean baseline, document the change, and confirm that a rollback also handles generated files, cached dependencies, credentials, and scheduled work.

External tool servers deserve the same questions

The official integrations documentation describes MCP as a way to connect Hermes to external tool servers. Whether the extension arrives as a plugin, an MCP server, or another supported interface, review the destination and capability boundary before treating it as a local operation.

Do not infer A loopback endpoint, a friendly catalog entry, or a skill name does not by itself prove local processing, harmless permissions, trusted provenance, or reversible behavior.

Verify the details where they change

These are direct Nous Research documentation destinations checked for this guide. Use the linked pages for current installation, configuration, security, catalog, and platform details rather than copying stale examples from a secondary page.

  • Hermes Agent documentation home ↗Orientation, installation destinations, platform links, and the official feature map.
  • Tools & Toolsets ↗Tool categories, toolset configuration, and the current built-in registry references.
  • Skills System ↗Skill locations, sources, catalog behavior, and security scanning guidance.
  • Plugins ↗Plugin discovery, capabilities, dependencies, consent, and management details.
  • Security ↗Approval, isolation, authorization, credential, and production-boundary guidance.