torchdrug — Claude Code skill: what it does, how to install
Helps build and debug TorchDrug 0.2.1 pipelines for molecular graphs and proteins. Use it when training property prediction, molecule generation, or retrosynthesis models.
npx skills add K-Dense-AI/scientific-agent-skills --skill torchdrug -a claude-codeWhat torchdrug does
The skill guides coding agents through the TorchDrug workflow stack: loading data via datasets.*, choosing representations like models.GIN, configuring objectives with tasks.*, and training through core.Engine. It covers molecular property prediction on ClinTox, autoregressive generation with GraphAF on ZINC250k, and retrosynthesis on USPTO50k.
Stack requirements are strict. TorchDrug 0.2.1 officially supports Python 3.7 to 3.10 and PyTorch 1.8 to 2.0. Apple Silicon is restricted to CPU-only execution because MPS lacks support, requiring source builds for torch-scatter and torch-cluster.
Environment setups need NumPy 1.x to prevent binary ABI clashes, alongside fair-esm and a single RDKit package. Attempting to mix rdkit and rdkit-pypi breaks the interpreter environment immediately.
What it helps with
- Configure molecular property prediction workflows with GIN and Engine
- Set up molecule generation models using GCPN or GraphAF
- Prepare USPTO50k datasets for retrosynthesis reaction center identification
- Resolve binary dependency matrix issues for PyTorch, torch-scatter, and RDKit
Example prompts for the agent
- Build a TorchDrug property prediction pipeline for ClinTox using GIN
- Configure GraphAF molecular generation on the ZINC250k dataset
- Check my virtualenv compatibility for TorchDrug 0.2.1 and torch-scatter
How to install torchdrug
1Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill torchdrug -a claude-code2Manually: put the files in place
~/.claude/skills/torchdrug/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/torchdrug/
1Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill torchdrug -a codex2Manually: put the files in place
.agents/skills/torchdrug/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill torchdrug -a cursor2Manually: put the files in place
.cursor/skills/torchdrug/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install K-Dense-AI/scientific-agent-skills torchdrug2Manually: put the files in place
.github/skills/torchdrug/SKILL.md1Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill torchdrug -a gemini-cli2Platform CLI
gemini skills install https://github.com/K-Dense-AI/scientific-agent-skills --path skills/torchdrugManually: put the files in place
.devin/skills/torchdrug/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill torchdrug -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
From the same set120
The whole set →Similar
Questions about torchdrug
How to install torchdrug in Claude Code?
Run npx skills add K-Dense-AI/scientific-agent-skills --skill torchdrug -a claude-code. The skill file installs to ~/.claude/skills/torchdrug/SKILL.md.
What does the torchdrug skill do?
It equips your AI assistant with validated patterns for TorchDrug 0.2.1, covering datasets, models, training engines, and dependency setups.
Does TorchDrug 0.2.1 support Apple Silicon GPU acceleration?
No, MPS is unsupported. On Apple Silicon machines, execution runs strictly on CPU, and libraries like torch-scatter must be built from source.