What it does

Scientific Agent Skills is a collection of 142 markdown skill files and database wrappers for the open Agent Skills standard. Install them with npx skills add or gh skill install, and your coding agent suddenly knows how to run Scanpy for single-cell analysis, query UniProt, or draft a literature review. It is essentially very good documentation that the agent can search and cite.

The interesting bit

The project treats “prompt engineering” as infrastructure. Each skill is a SKILL.md with curated examples, best practices, and integration guides — the kind of thing a postdoc would scribble in a notebook, but structured so an LLM can follow it reliably. The breadth is the point: 78 public databases, 70+ Python packages, and even lab-automation APIs like Opentrons.

Key highlights

• 142 skills covering genomics, cheminformatics, proteomics, clinical research, materials science, and more
• Unified database-lookup skill for PubChem, ChEMBL, COSMIC, ClinicalTrials.gov, FRED, USPTO, and 70+ others
• Pre-documented Python package skills for RDKit, scVelo, OpenMM, PennyLane, TimesFM, and others
• Integration skills for Benchling, DNAnexus, OMERO, Protocols.io, and lab hardware
• Works with Cursor, Claude Code, Codex, Google Antigravity, and any Agent Skills-compatible agent

Caveats

• The README is vague on what “tested, validated examples” actually means; no benchmarks or validation studies are shown
• The “160,000+ scientists worldwide” claim in the repo description is unsupported in the README itself
• Skills are documentation and wrappers, not executable code — your agent still needs the underlying Python packages and API keys installed

Verdict

Grab this if you are a biologist or chemist using Cursor/Claude and tired of explaining BLAST or scRNA-seq preprocessing to your agent for the tenth time. Skip it if you are looking for a standalone scientific computing library; this is glue, not engine.