GPTomics/bioSkills

bioSkills – AI‑agent “skills” for bioinformatics

What it is – bioSkills is a curated library of “skills” (code snippets, best‑practice templates and usage guides) that teach large‑language‑model coding agents (Claude Code, OpenAI Codex, Google Gemini/Antigravity, OpenCode, OpenClaw) how to write correct, idiomatic bio‑informatics scripts. Each skill covers a concrete task – e.g. reading FASTA, running a variant‑calling pipeline, or performing single‑cell differential expression – and bundles the preferred tools, command‑line flags and example code.

Why it matters – Modern LLM‑based coding assistants can generate code, but they often miss domain‑specific nuances (file formats, tool versions, statistical conventions). By loading bioSkills into an agent, the model gets a ready‑made knowledge base that improves the accuracy and reproducibility of the code it suggests for computational biology workflows.

Key content

• 41 skill categories ranging from low‑level sequence I/O to high‑level multi‑omics integration, chemoinformatics, liquid‑biopsy, and workflow management.
• Tool coverage includes the most widely used Python, R/Bioconductor and CLI utilities (e.g. Biopython, samtools, DESeq2, Seurat, GATK, AlphaFold, RDKit, Snakemake, Nextflow, etc.).
• Examples & references – each skill ships an examples/ folder with runnable scripts and a usage‑guide.md that explains the rationale behind each command‑line option.
• Evaluation – a benchmark report (linked in the repo) shows the skills’ impact on the Bio‑Task Bench dataset, with a performance plot included.

Installation – The repo provides thin wrapper scripts for each supported LLM platform. For example, to make the skills available to Claude Code globally:

git clone https://github.com/GPTomics/bioSkills.git
cd bioSkills
./install-claude.sh          # global install
# or install into a specific project
./install-claude.sh --project /path/to/my/project
# install only the single‑cell and variant‑calling categories
./install-claude.sh --categories "single-cell,variant-calling"

Similar install‑codex.sh, install‑antigravity.sh, install‑opencode.sh and install‑openclaw.sh scripts exist, all supporting --list, --validate, --update, --dry‑run and --uninstall flags.

How to use – After installation the agent’s skill directory (e.g. ~/.claude/skills/ or ~/.agents/skills/) contains a hierarchy like:

sequence-io/
  read_fasta.py
  write_fastq.py
  usage-guide.md
variant-calling/
  germline_gatk.py
  sv_manta.py
  usage-guide.md
…

When you prompt the LLM to "write a pipeline that calls germline SNPs from a BAM file", the model can retrieve the germline_gatk.py template, adapt paths, and emit a ready‑to‑run script that respects best‑practice flags (read groups, VQSR, etc.). The same works for R‑based tasks – e.g. a Seurat single‑cell clustering skill that returns a reproducible R script with proper QC thresholds.

Dependencies – The README lists the required Python packages, R/Bioconductor libraries and system tools (samtools, bcftools, blast, minimap2, etc.). Installation scripts do not install these external tools; users must satisfy them beforehand (via pip, conda, brew or apt).

Status – The repository is archived (no further updates), but the structure is still valid. The authors encourage users to fork the repo and customise or extend the skill set for their own domains.

---

TL;DR – bioSkills is a comprehensive, pre‑packaged knowledge base that equips LLM coding agents with domain‑specific bioinformatics expertise, enabling them to generate accurate, reproducible analysis scripts across a huge range of genomics, proteomics, and chemoinformatics tasks.