Skip to content
Get packs

Guide · Agent skill

How to use the Hugging Science Catalog Finder agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The hugging-science skill teaches an AI agent to search the Hugging Science catalog, a curated index of scientific machine learning datasets, models, blog posts and demos, and to pick a suitable resource for a task. This guide is for people who may never have used agent skills before. By the end you will know what the skill does, how to install it, how to ask for it in plain language, and what to check before relying on anything it finds.

Hugging Science Catalog FinderK-Dense-AI/scientific-agent-skills · MIT

Quick start

  1. 1Download the ZIP. Get the hugging-science ZIP from the page where you found the skill.
  2. 2Install it. Unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
  3. 3Make a request. Describe a scientific task in plain language, naming the field, data type and goal.
  4. 4Review the candidates. Read the trade-offs the agent gives between datasets, models or Spaces, and ask about anything unclear.
  5. 5Verify the licence and commit. Open the model card or dataset page, confirm the licence, and note the exact commit before you use it.
  6. 6Load the resource. Follow the loading notes for that resource type, and set HF_TOKEN in .env if the resource is gated.

What an agent skill is and what this one does

An agent skill is a folder of instructions and optional scripts that an AI agent loads when a task matches it. The entry file, SKILL.md, tells the agent when to use the skill and how to work through the task.

The hugging-science skill covers AI and machine learning applied to science. It names 17 topic areas, including biology, chemistry, genomics, materials science, climate, physics, astronomy, medicine, mathematics and protein design. For a request in one of these areas, the agent fetches catalog entries, compares the candidates, and follows reference notes on how to load each one.

  • Finds candidate datasets, models, blog posts and Spaces (interactive demos) for a scientific task.
  • Compares options on scale fit, licence and access, modality (for example DNA, protein, SMILES or crystal structure) and recency.
  • Points to the reference notes for loading each type of resource.
  • Reminds the agent to pin exact versions and verify licences and model cards.

Install the hugging-science skill

Download the ZIP for the skill from the page where you found it. Then choose one of the options below.

For Claude Code, unzip the download into ~/.claude/skills/ to make it available to all your projects, or into a project's .claude/skills/ folder to limit it to that project.

In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

  • Personal install: ~/.claude/skills/ (available everywhere).
  • Project install: .claude/skills/ inside your project folder.
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP.

Ask for it with natural requests

You do not need to name the skill. The skill is meant to engage when your task involves AI or ML applied to science and you mention a scientific domain, a known scientific model, or a need for a demo. Generic ML requests are outside its scope.

Good triggers look like the examples below. Include the field, the data type and your goal, since the agent weighs scale, modality and licence against them.

  • Is there an open protein language model I can fine-tune on a small dataset?
  • Find scientific datasets for single-cell analysis and tell me their splits and licences.
  • Compare Evo-2 and ESM2 for a sequence classification task on a laptop.
  • Find a Gradio demo for protein binder design.
  • Which climate datasets on the catalog could I use to benchmark a forecasting model?

How the agent searches the catalog

The skill uses a bundled Python script, scripts/fetch_catalog.py, to read catalog content in a structured way. It supports topic lookups, filtering by type and keyword search. You can also run these commands yourself to see what the agent will see.

The catalog also publishes plain markdown files at huggingscience.co, including llms.txt (a compact index), llms-full.txt (every entry) and one file per topic. The agent can read these directly if needed.

python scripts/fetch_catalog.py topic biology python scripts/fetch_catalog.py topic materials-science --filter models python scripts/fetch_catalog.py search "protein language model" python scripts/fetch_catalog.py all

Verify a resource before you use it

Being in the catalog does not establish quality or openness. The skill asks the agent to check the underlying Hugging Face model card, the licence, and the author documentation for every resource it recommends. Before use, the agent records the exact repository, the immutable commit, the dataset configuration and split, the licence, and the tokenizer or preprocessing revision.

Some resources need a specific runtime. For example, the skill notes that Evo2 requires its native tooling rather than the standard Transformers loader. The catalog provides discovery; each resource is used through its verified loader or service.

  • Pin the exact commit, not just a branch name.
  • Check the licence and model card yourself.
  • Confirm the input modality and normalization match your data.
  • If the catalog does not list a resource, search the Hugging Face Hub directly.

Requirements, safety and network use

The skill requires Python 3.10 or newer and network access to huggingscience.co for the catalog fetcher. Running particular scientific models needs the matching scientific runtime. Some Hugging Face resources are gated; for those you can add an HF_TOKEN value to a .env file. The skill prefers .env over logging in through the command line, and it says not to hard-code or echo tokens.

This skill runs scripts and uses the network. Read the bundled scripts before installing a skill from any source you do not trust. The fetcher is described as standard-library only, and the skill guides discovery and usage but does not run scientific models for you.

  • Python 3.10 or newer.
  • Network access to huggingscience.co.
  • Matching scientific runtime for some models, such as native Evo2 tooling.
  • Optional HF_TOKEN in .env for gated resources; many resources are public and work without it.

Tips for better results

Give the agent the field, the modality, your hardware limits and whether you plan to fine-tune or only evaluate. These decide between candidates more than keyword overlap does.

If you are starting in a field, ask for the flagship picks first. The skill keeps a short list of go-to resources per domain and asks the agent to confirm them against the live catalog, since entries change.

  • State your hardware limits so the agent can judge model size.
  • Ask for the methodology blog post behind a model when you want to know why it was designed that way.
  • Ask the agent to list the licence and the exact commit it will use.

What people use it for

  • Choosing a starting protein or DNA model for a fine-tuning project in biology.
  • Finding single-cell or genomics datasets with clear splits before an evaluation.
  • Comparing a small and a large scientific model to fit a laptop or a workstation.
  • Locating a Gradio demo for a scientific task such as binder design.
  • Gathering the methodology posts behind a model to cite in a report.

Frequently asked questions

Is the hugging-science skill free and what licence does it use?

The skill is published under the MIT licence. The catalog lists resources that each carry their own licence, so check the licence on the specific model or dataset before you use it.

What do I need to run the hugging-science skill?

You need Python 3.10 or newer and network access to huggingscience.co for the catalog fetcher. Some models need a matching scientific runtime, and gated Hugging Face resources need an optional HF_TOKEN in a .env file.

How do I install the hugging-science skill?

Download the ZIP and unzip it into ~/.claude/skills/ for all your projects, or into a project's .claude/skills/ folder. In the Claude apps, you can upload the ZIP under Settings → Capabilities → Skills.

Does the skill run scientific models for me?

No. The skill guides discovery and usage. It helps the agent find and compare resources and explains how to load them, but it does not run the scientific models itself.

Is this skill a good alternative to searching Hugging Face directly?

It is useful when you want a curated, domain-organised starting point. The catalog is not exhaustive, so if a resource is missing, the skill itself advises searching the Hugging Face Hub directly.

Why does the agent say a catalog listing is not proof of quality?

Curation selects candidates but does not guarantee quality, openness or that a resource runs as described. The skill asks the agent to check each model card, licence and author documentation before recommending or using a resource.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are a researcher, ML engineer or student who needs to find and compare open scientific datasets, models or demos in a specific field. Pick something else if you need generic ML help such as chatbots or recommendation systems, or if you expect the skill to run scientific models for you.

Download the skill