nvMolKit skill
A skill that teaches an agent to write nvMolKit code: GPU versions of RDKit fingerprints, similarity, conformers and clustering.
- 282 ★ on the repository
- Last commit
Minimal access
Why this level
- The skill holds only instructions and recipes; limits mostly concern driver and CUDA compatibility.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/nvmolkitDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill holds only instructions and recipes; limits mostly concern driver and CUDA compatibility.
This is third-party code. Review the repository files before installing.
What it does
The `nvmolkit-usage` skill from the NVIDIA repository gives an agent a map of the public nvMolKit Python interface and ready recipes. nvMolKit reproduces batched RDKit operations on a GPU: Morgan fingerprints, Tanimoto similarity, ETKDG conformers, MMFF and UFF optimization, clustering. Inputs and outputs remain real `rdkit.Chem.Mol` objects. The `skills/nvmolkit-usage/` folder holds `SKILL.md`, an `advanced-usage.md` reference and `evals` test sets. The skill also says when a GPU pays off: large batches of identical operations where total throughput matters.
Who it is for. Cheminformatics specialists and developers asking an agent to speed up screening of large molecule libraries on an NVIDIA GPU.
Good fit when
- A library of tens of thousands of molecules waits for fingerprinting.
- All-pairs similarity over a collection takes too long on a CPU.
- A batch of molecules needs conformers and energy minimization.
- An agent writes nvMolKit code and confuses function names.
Not a fit when
- You have one molecule and latency matters more than throughput.
- The machine has no NVIDIA GPU of compute capability 7.0 or higher.
- You need to build nvMolKit from source: the skill does not cover it.
Example request
Compute Morgan fingerprints for a list of SMILES on the GPU and find Tanimoto similarity to a query. Use nvMolKit and explain how to pass the result on.Limitations
nvMolKit itself runs only on NVIDIA GPUs of compute capability 7.0 or newer, and the driver must support CUDA 12.6 or later. You need a CUDA build of PyTorch matching the driver: pip may pick an incompatible one, so install `torch` beforehand. The skill covers the installed Python interface and does not explain building from source. A single molecule gains nothing from a GPU. For several GPUs and large batches read the `advanced-usage.md` reference: the main skill file holds only basic recipes.
How to disable. Delete the `nvmolkit-usage` folder from the project's `.cursor/skills/` or from `~/.cursor/skills/`. The nvMolKit package itself is removed separately through the package manager.
FAQ
Where do I copy the skill?
Into the project's `.cursor/skills/` to use it in one project, or into `~/.cursor/skills/` for every project. The agent picks it up on terms such as `nvmolkit` and `Tanimoto`.
Which package manager should I use for nvMolKit?
The recommended README option is conda-forge with the `pytorch-gpu` metapackage. With pip install `torch` for the right CUDA first, with uv add `--torch-backend`.
Does the skill work in Claude Code?
The documentation names Cursor and other agents supporting the `SKILL.md` format. For other clients put the folder into their skills directory.
What do the `evals` sets check?
They hold evaluation prompt definitions and skill-trigger checks. They help confirm the agent picks the skill up and writes working code.
Which operations does the skill cover?
Morgan fingerprints, Tanimoto similarity, ETKDG conformers, MMFF and UFF optimization, clustering and molecule search. Skill triggers include these terms.
Discussion
No questions or reviews yet.
Related
SkillsA global intelligence MCP server: news, geopolitical risk, markets and infrastructure exposed as agent tools
166 skills for scientific work: bioinformatics, cheminformatics, clinical data, geospatial analysis and 100+ databases
The native ClickHouse client executes SQL and exports results for agent-prepared analytical queries reviewed by a user.
Streamlit agent skill
developing-with-streamlit
The official Streamlit skill bundled with the library: building, styling, debugging and speeding up apps, installed with streamlit skills
Two skills from Stas Bekman's book: training large models on a cluster and an acceptance run of a GPU cluster with a report.
migrate CLI applies versioned database changes in order, with a chosen step count or target schema version for each run.
README in short
The repository holds the nvMolKit library itself and a skill with an interface map, recipes and checks. Library installation and skill copying are described separately.