142 cheat sheets that turn coding agents into scientists

What it does

Scientific Agent Skills is a collection of 142 curated “skills” — essentially documentation bundles with code examples, best practices, and integration guides formatted for the open Agent Skills standard. Instead of asking an AI agent to blindly guess how to run RNA velocity analysis or query PubChem, you feed it a pre-vetted skill file that covers workflows across genomics, cheminformatics, proteomics, and clinical research. The agent still writes Python as usual, but now it has explicit, maintained context for packages like RDKit, Scanpy, and OpenMM rather than relying on generic training data.

The interesting bit

The project treats prompt context like a standard library: each skill is a standalone module with a SKILL.md, use cases, and working examples, so agents can discover and invoke them automatically. It is largely glue code and documentation architecture, but the value is in the curation — someone already mapped 100+ databases and 70+ Python packages to agent-friendly interfaces.

Key highlights

• 142 skills covering everything from single-cell RNA-seq and molecular docking to grant writing and regulatory compliance
• Unified database access to 78 public databases including PubChem, ChEMBL, UniProt, and ClinicalTrials.gov, plus multi-database wrappers like BioServices and gget
• Explicitly optimized skills for 70+ Python packages including PyTorch Lightning, PennyLane, Qiskit, and MDAnalysis
• Integration skills for lab platforms like Benchling, DNAnexus, OMERO, and Opentrons
• Works with any agent supporting the open Agent Skills standard, including Cursor, Claude Code, Codex, and Google Antigravity

Caveats

• The README concedes that agents can already use any Python package or API on their own; these skills only make them “stronger and more reliable,” so the value depends on how much you trust your agent’s baseline
• The 100+ database count aggregates multi-database packages like BioServices (~40 services) and BioPython (38 NCBI sub-databases), so the headline figure is fuzzier than it sounds

Verdict

Researchers and bioinformaticians who already use AI coding agents and are tired of babysitting them through API docs will get the most value; if you do not use agent-based coding tools, this is just a very well-organized collection of markdown recipes you cannot directly use.