NousResearch just announced that Hermes Agent’s creative-ideation skill is the real deal, and it works way better than asking your AI to be creative and hoping for the best.
Instead of generic brainstorming that sounds like a college freshman’s first attempt at being profound, this skill routes your prompt through one of 22 named creative methodologies from actual artists and thinkers. Oulipo, SCAMPER, TRIZ, Oblique Strategies, first principles, biomimicry — they’re all in there, each with a reference file that tells the agent exactly how to apply the method.
The tweet shows a video of the skill in action inside the Hermes TUI. The user types something like I need a fresh angle for this project and the agent routes through the library, picks the right method, and generates non-obvious output instead of generic LLM-mode ideas.
What the skill actually does
The creative-ideation skill is not a single method. It is a router with a library of 22 named techniques, each grounded in real creative practice. The skill’s design is explicit about when to use which method:
- Generating ideas from scratch with no constraint? It hits the constraint library and dispatches structure.
- Expanding a half-formed concept? SCAMPER or Oblique Strategies.
- Selecting between options? Premortem, inversion, or compression progress.
- Stuck or stale on a problem? Oblique Strategies, lateral provocations, pataphysics.
- Refining text or copy? Defamiliarization.
- Synthesizing a pile of notes or research? Affinity diagrams.
The routing is phase-first, domain-second. The 4-step procedure extracts three signals from the prompt — phase, domain, and specificity level — before it even touches the method catalog. That structure is what prevents creative-ideation from being another here are 5 generic ideas generator.
v2.2.0 just made the router smarter
The new version (still in PR at #50344 as of June 21) upgrades the routing layer to a full diagnostic model:
\[intent -> bottleneck -> creative move -> prescribe\]
Instead of matching your prompt against tables, the router now asks:
- Intent — What are you actually trying to do? Not just the surface request, but the underlying need.
- Bottleneck — What is the one binding constraint? No constraint? No direction? Frame-locked? Disconnected from your material?
- Creative move — What kind of novelty closes that gap?
- Prescribe — One method, or a principled stack.
This is paired with Margaret Boden’s three-creativity taxonomy from her book The Creative Mind:
- Combinational — new links between existing ideas. Mix and match what is already known.
- Exploratory — search the conceptual space. Push deeper into the established terrain.
- Transformational — change the space itself. Break the generative constraints.
The taxonomy gives the router a principled way to decide what kind of creativity a situation actually needs. The skill explicitly says that slop — the safe center of the conceptual space — is what the default LLM produces, and the whole point of the diagnostic router is to make that harder to output by default.
Why this matters for Hermes Agent users
If you already use Hermes Agent — and I do, every day — the skill system is the part that keeps getting more interesting. The /learn command turned any source material into a reusable skill. v0.17.0 shipped 1,475 commits and the Reach release across new channels.
Creative-ideation is the next layer. The skill system turned Hermes into an agent that can learn from anything. This skill turns it into an agent that can think with the same generative frameworks real artists and designers use.
The full library includes references for each method with primary-source attribution. Oblique Strategies is Brian Eno’s deck. Oulipo is Raymond Queneau’s constrained writing movement. SCAMPER is Bob Eberle’s creative thinking technique. TRIZ is Genrich Altshuller’s engineering invention framework. The skill does not pretend these are original — it cites them and lets the agent apply them.
Where it still needs a human
I have been running creative-ideation in my Hermes setup for a while (it is installed by default in the current profile system). The routing works, but it is not magic. The skill’s own rules say to refuse the first three ideas, because they are slop. The diagnostic router can identify the bottleneck, but it cannot know your actual material the way you do.
For generating starting points when you are stuck? Excellent. For expanding a direction that is too safe? Yes. For replacing the actual work of creative thinking? No, and the skill does not claim to. The attribution lines in each output name the method and who invented it, which is a nice check against the agent pretending it came up with something on its own.
The most honest part of the skill’s documentation is the anti-default check: if you are about to write a bare numbered list, stop and pick a method first. The router is designed to prevent what every other generator does.
Bottom line
The creative-ideation skill is the most thoughtful implementation of AI-assisted creative thinking I have seen in an agent product. It does not pretend to be creative itself. It routes to real methods, cites real practitioners, and comes with a diagnostic router that is getting smarter with each version.
If you write, design, build products, or do any work where generating options is half the battle, this skill is worth trying on its own. The fact that it lives inside an agent that also codes, researches, and manages your workflow is the bonus.
The v2.2.0 diagnostic router is still an open PR, but the v2.1.0 method library is live in Hermes Agent now. Go open the Hermes TUI, load the skill, and give it a prompt you would normally struggle to brainstorm. The first three ideas it gives you will be wrong. The fourth one might be something you would not have thought of.
