ClawBio

ClawBio is a bioinformatics-native AI agent skill library built around a simple idea: skills should be reproducible specifications, not stochastic LLM output. It ships 105 skills covering pharmacogenomics reporting, GWAS lookup, ancestry PCA, polygenic risk scoring, UK Biobank navigation, and drug photo identification, each with demo data and reproducibility bundles. Everything runs locally by default via pip install clawbio, and skills also work as Claude Code plugins or plain Agent Skills folders. The project reports 1,160+ GitHub stars, benchmark validation against curated ground truth, and a fully open human genome (the Corpasome, CC0) as demo data.

The trade-off is scope: this is a platform with its own conventions (reproducibility bundles, skill scaffolding) rather than a single drop-in skill, so expect a learning curve. Some skills reach out to public annotation APIs, which the project documents per skill, so privacy-sensitive users should check the data-handling docs.

Install via pip: pip install clawbio, then run a skill in demo mode with clawbio run pharmgx --demo. For Claude Code, add the marketplace with /plugin marketplace add ClawBio/ClawBio. For other Agent Skills-compatible editors, copy or symlink skill folders from skills/ into ~/.agents/skills/.