🧩 Hermes Agent Skills Hub: Real CLI Tools, Not Prompts
Why I'm Writing This
I use TouchDesigner — a nod-based environment from Derivative where you build everything from nodes: CHOPs for data, SOPs for geometry, TOPs for textures, COMPs for components. You can do generative art, data visualization, VJ sets, interactive installations, real-time video processing. Download it free — the non-commercial license covers most use cases.
I have NewProject1.toe open on my desktop right now. datto1 CHOP, hessen_geo geoCOMP holding a dot Sphere SOP, render1 and cam1 — a basic scene I spin up when testing ideas.
I also use Hermes Agent — an AI agent from Nous Research that works through CLI and recently got a desktop version on Electron. I give it tasks, it executes them through terminal, browser, API. It has skills — extensions that add new tools.
I realized something: TouchDesigner can visualize literally anything. Feed it data — it draws. Feed it text — it renders. Feed it a 3D model — it shows it. The only question is how to get and process that data.
Hermes can process text, code, web data, run scripts. But it can't draw. TouchDesigner can draw, but it can't think.
So I thought: what if I connect them? Through MCP (Model Context Protocol). So Hermes can send data to TouchDesigner, TouchDesigner can respond, and the whole thing works as one organism.
This isn't just a nice idea. It's a practical need: I want my AI agent to visualize its work in real-time, right in TD. And I want to control TD from my Hermes chat.
That's when Skills Hub appeared.
What Skills Hub Actually Is
Nous Research launched Skills Hub — a catalog of skills for Hermes Agent. I opened it and realized: this isn't a prompt catalog like I initially thought. These are real CLI tools you can install and invoke directly from a session.
Not "ask the agent to do something." Install a skill — and you get a new command.
The Numbers Are Messy (I Want to Be Honest)
The original tweet from @Fluyeporlaweb says "89 built-in, 81 optional, 521 community." My own screenshot shows different numbers: 74 built-in, 95 optional, and 90,576 community entries (including third-party registries). The discrepancy probably means the tweet was posted after the screenshot was taken — the hub refreshes twice daily. I'm going with ~691 as the working estimate.
| Source | What | Screenshot | Tweet |
|---|---|---|---|
| Built-in | Ships with Hermes | 74 | 89 |
| Optional | Install separately | 95 | 81 |
| Community | All registries | 90,576 (all) | 521 (curated) |
| Registries | Data sources | Built-in + Optional + Anthropic + OpenAI + HuggingFace + NVIDIA + skills.sh (19,956) + ClawHub (69,471) + browse.sh (377) + LobeHub (505) + gstack (52) |
What These Skills Actually Do
Here are concrete examples I found on the first page:
macos-computer-use
Takes desktop screenshots, moves mouse, types keys, scrolls. Runs in background without stealing focus. This is not "instructions for controlling macOS." It's a real headless UI driver. I've tested it — it actually clicks on my screen while I work in another window.
comfyui
Generates images, video, and audio via ComfyUI from inside a Hermes session. Not "ask ComfyUI nicely" — actual pipe calls through CLI. I've installed it, it works.
humanizer
Strips passive voice, AI clichés, and generic phrasing from text. Returns text that doesn't look generated. I've used it for Telegram posts.
popular-web-designs
54 real design systems (Stripe, Linear, Vercel, etc.) as deployable HTML/CSS templates. Not links to articles — actual code.
claude-code / codex
Delegates coding tasks to Claude Code CLI and OpenAI Codex CLI respectively. Hermes doesn't try to write the code itself — it hands the task to the right specialist.
manim-video
3Blue1Brown-style math/algorithm visualizations. These are Python scripts that render video.
excalidraw
Hand-drawn architecture diagrams from text descriptions. I've used it for post diagrams.
ascii-video
Converts any video to colored ASCII MP4 or GIF.
touchdesigner-mcp
And this is the most interesting one for me. Controls a running TouchDesigner instance through MCP. Two channels: one for sending data to TD, one for receiving. This is exactly what I was thinking about at the start of this article.
The Compatibility Part That Matters
These aren't Hermes-only plugins. They work through the standard MCP (Model Context Protocol), meaning you can also install them in:
- Claude Code
- Cursor
- Codex
- OpenCode
- Any MCP-compatible agent
Hermes is the coordinator, not the sole consumer.
What Changed for My TouchDesigner Workflow
Before: I wrote a Python script outside TD, ran it, then manually moved the result into TD through DAT or CHOP.
Now: I can tell Hermes "visualize this API data in TD", Hermes installs the right skills (or uses existing ones), data flows through MCP directly into TD — into a CHOP or TOP — and TD draws in real-time.
Or the reverse: TD sends data (e.g., audio spectrum) through MCP to Hermes, Hermes analyzes and responds.
Price
Zero. Open source. No paywall, no token credits per call. The Hermes repo has 186k stars and 31.9k forks.
The Hub Website Is Down
The Skills Hub website (hermes-agent.nousresearch.com/skills) is currently returning an error. I checked myself — right now. But the repository is fully active and hermes skills install works from CLI. The website is a wrapper, not the product itself.
Bottom Line
Skills Hub makes Hermes not just a "chatbot with terminal access" but a platform with an ecosystem. And for me, this is concrete: I can now connect my TouchDesigner with an AI agent through a standard protocol, without writing a single line of integration code.
This is not a "prompt marketplace." This is a toolkit I can use right now.
Source: X/Twitter post by @Fluyeporlaweb
TouchDesigner: derivative.ca | Download
Skills Hub: hermes-agent.nousresearch.com/skills (site temporarily down)
Hermes Docs: hermes-agent.nousresearch.com/docs
Hermes Desktop: hermes-agent.nousresearch.com/desktop
GitHub: github.com/NousResearch/hermes-agent
My Telegram: t.me/azamat_dasein
Top comments (0)