SciAgent-Skills

Turn your AI coding agent into a life sciences expert — 199 bioinformatics skills for Claude Code covering RNA-seq, single-cell analysis, genomics, proteomics, drug discovery, and more. Boosted BixBench from 65% to 92%. Open source.

SciAgent-Skills is the largest open-source skill library for scientific AI agents. It equips Claude Code (and any markdown-compatible agent) with domain-specific knowledge for computational biology, bioinformatics, cheminformatics, and biostatistics — no fine-tuning required, just plug in and analyze.

Keywords: bioinformatics AI agent, Claude Code skills, scientific computing, RNA-seq analysis, single-cell RNA-seq, drug discovery pipeline, protein structure prediction, computational biology tools, life science automation, BixBench benchmark

Benchmark: 92.0% on BixBench-Verified-50

BixBench is a benchmark for evaluating AI agents on real-world bioinformatics tasks. SciAgent-Skills achieved 92.0% accuracy on BixBench-Verified-50, the highest among all tested systems:

Simply equipping Claude Code with these domain-specific skills yields a +26.7 percentage point improvement — no fine-tuning, no custom model, just structured scientific knowledge.

Try It Now — OmicsHorizon

Want to try these skills without any setup? OmicsHorizon (오믹스 호라이즌) is the web platform powered by SciAgent-Skills. Sign up and start analyzing your bioinformatics data directly in your browser — RNA-seq, proteomics, drug screening, and more.

199 ready-to-use scientific skills for AI coding agents — covering genomics, proteomics, drug discovery, biostatistics, scientific computing, and scientific writing.

Each skill is a self-contained SKILL.md file with runnable code examples, key parameters, troubleshooting guides, and best practices. Designed for Claude Code, but compatible with any AI agent that reads markdown skill files (setup guides below).

What's Inside

Skill types: 72 toolkits, 53 database connectors, 37 guides, 37 pipelines

Installation

Prerequisites

• An AI coding agent: Claude Code, OpenAI Codex CLI, Cursor, or Windsurf
• Git
• Python 3.12+ (only needed if you want to run validation scripts)

Note: SciAgent-Skills is not an npm package. Skills are plain markdown files read directly by your AI agent — no npx, npm install, or runtime dependencies needed. Just clone the repository and point your agent at the skill files.

Step 1: Clone the Repository

git clone https://github.com/jaechang-hits/SciAgent-Skills.git
cd SciAgent-Skills

Step 2: Choose Your Setup Method

Method A: Claude Code Plugin (Recommended)

Load SciAgent-Skills as a Claude Code plugin for the current session:

claude --plugin-dir /path/to/SciAgent-Skills

To verify the plugin loaded, run /plugin inside Claude Code and check that sciagent-skills appears in the Installed tab.

Skills become available as /sciagent-skills:<skill-name>:

/sciagent-skills:scanpy-scrna-seq
/sciagent-skills:rdkit-cheminformatics
/sciagent-skills:pymc-bayesian-modeling

Or just describe your task — the agent finds the relevant skill automatically:

"Perform differential expression analysis on this RNA-seq count matrix"

Persistent installation — to load the plugin automatically in every session, use the plugin install command inside Claude Code:

/plugin marketplace add jaechang-hits/SciAgent-Skills
/plugin install sciagent-skills

Method B: Project-Level Integration (Claude Code)

Clone into your project directory so Claude Code picks up skills via CLAUDE.md:

cd your-project
git clone https://github.com/jaechang-hits/SciAgent-Skills.git .sciagent-skills

Add to your project's CLAUDE.md:

## Scientific Skills
Reference skills in `.sciagent-skills/skills/` for domain-specific analysis.
Registry: `.sciagent-skills/registry.yaml`

Using with Other Agents

SciAgent-Skills works with any AI agent that can read project files. Clone the repo into your project, then configure the agent to discover skills via registry.yaml.

Method C: OpenAI Codex CLI

cd your-project
git clone https://github.com/jaechang-hits/SciAgent-Skills.git .sciagent-skills
cp .sciagent-skills/integration-templates/AGENTS.md ./AGENTS.md

Codex reads AGENTS.md at the project root automatically. If you already have an AGENTS.md, append the contents from the template.

Method D: Cursor

cd your-project
git clone https://github.com/jaechang-hits/SciAgent-Skills.git .sciagent-skills
mkdir -p .cursor/rules
cp .sciagent-skills/integration-templates/cursor-rules.md .cursor/rules/sciagent-skills.md

Cursor loads rules from .cursor/rules/ automatically. Alternatively, you can use the AGENTS.md template — Cursor supports it as well.

Method E: Windsurf

cd your-project
git clone https://github.com/jaechang-hits/SciAgent-Skills.git .sciagent-skills
mkdir -p .windsurf/rules
cp .sciagent-skills/integration-templates/windsurf-rules.md .windsurf/rules/sciagent-skills.md

Windsurf loads rules from .windsurf/rules/ automatically. Alternatively, you can use the AGENTS.md template — Windsurf supports it as well.

Other agents: If your agent reads project files, clone the repo as .sciagent-skills/ and instruct the agent (via its config mechanism) to read .sciagent-skills/registry.yaml for skill discovery.

Step 3: Install Dependencies

cd SciAgent-Skills
pixi install

Pixi handles the Python environment and all required packages. If you don't have pixi installed:

curl -fsSL https://pixi.sh/install.sh | bash

How Skills Work

Each skill follows a structured template:

skills/<category>/<skill-name>/
  SKILL.md          # Main skill file (300-550 lines)
  references/       # Optional deep-dive reference files
  assets/           # Optional templates, configs

A SKILL.md contains:

• Frontmatter — name, description, license (for agent discovery)
• Overview & When to Use — what the tool does and when to reach for it
• Prerequisites — packages, data, environment setup
• Quick Start — minimal copy-paste example
• Workflow / Core API — step-by-step pipeline or module-by-module API guide
• Key Parameters — tunable settings with defaults and ranges
• Common Recipes — self-contained snippets for common tasks
• Troubleshooting — problem/cause/solution table

The agent reads only the description field during planning. Full skill content is loaded on demand when relevant.

Directory Structure

SciAgent-Skills/
├── .claude-plugin/
│   └── plugin.json        # Claude Code plugin manifest
├── integration-templates/  # Config templates for Codex, Cursor, Windsurf
├── skills/                 # All 199 skills organized by category
│   ├── genomics-bioinformatics/
│   ├── structural-biology-drug-discovery/
│   ├── scientific-computing/
│   ├── cell-biology/
│   ├── biostatistics/
│   ├── scientific-writing/
│   ├── systems-biology-multiomics/
│   ├── proteomics-protein-engineering/
│   ├── lab-automation/
│   ├── data-visualization/
│   └── molecular-biology/
├── templates/              # Skill authoring templates
├── registry.yaml           # Index of all skills
├── CLAUDE.md               # Skill authoring guide
└── scripts/
    └── validate_registry.py

Example Use Cases

Single-Cell RNA-seq Analysis (scRNA-seq)

"Load 10X data, QC filter, normalize, cluster, find marker genes, and annotate cell types"

Uses: anndata-data-structure → scanpy-scrna-seq

Bulk RNA-seq Differential Expression

"Run DESeq2-style differential expression analysis on th