Skip to content
/
Tier B A11 Governance

Claude Scientific Skills

claude-scientific-skills · K-Dense-AI/claude-scientific-skills · ★ 26k · last commit 2026-05-25

Primitive shape 138 total
Skills 138
00

Summary

claude-scientific-skills — Summary

claude-scientific-skills (now named "Scientific Agent Skills") is the largest skill collection in the corpus — 138 skills spanning biology, chemistry, medicine, physics, engineering, data analysis, geospatial science, laboratory automation, and scientific communication. Maintained by K-Dense Inc., it covers 100+ scientific databases (PubChem, ChEMBL, UniProt, ClinicalTrials.gov, FRED, USPTO), 70+ optimized Python package skills (RDKit, Scanpy, PyTorch Lightning, BioPython, OpenMM, etc.), and research communication tools (literature review, peer review, grant writing, poster/slide generation). With 26,062 stars and version 2.39.0, it is by far the highest-starred framework in this batch. The project has pivoted to support the open agentskills.io standard, making it portable across Cursor, Claude Code, Codex, and other agents. It also ships a security scanner (scan_skills.py) using cisco-ai-skill-scanner that audits all 138 skills for behavioral anomalies.

Differs from seeds: No seed approaches this skill breadth (138 vs superpowers' 14). Closest to superpowers (skills-only, no commands) but operates at a vertical domain depth that superpowers explicitly avoids. The agentskills.io portability standard and the integrated security scanner are features not present in any seed.

01

Overview

claude-scientific-skills — Overview

Origin

Created by K-Dense Inc. (a startup focused on scientific AI). The project has been renamed from "Claude Scientific Skills" to "Scientific Agent Skills" to reflect its portability across AI agents. Current version: 2.39.0.

Philosophy

"Transform your AI coding agent into a 'AI Scientist' on your desktop!"

"While the agent can use any Python package or API on its own, these explicitly defined skills provide curated documentation and examples that make it significantly stronger and more reliable for the workflows below."

The core insight is that AI agents can already call any Python package or API without a skill file, but the skill file provides:

  1. Curated documentation that reduces hallucinated API calls
  2. Best practices and common patterns
  3. Integration guides for scientific databases
  4. Performance-optimized code examples

The K-Dense BYOK Ecosystem

The repo now links to K-Dense BYOK — a free desktop app that bundles all 138 skills with:

  • Web search and file handling
  • 100+ scientific databases
  • 40+ model support (bring-your-own-key)
  • Optional cloud compute via Modal for heavy workloads

Open Standard

The project adopted the agentskills.io open standard, making it compatible with any agent that reads the skill format — not just Claude Code.

Security

scan_skills.py audits all 138 skills using cisco-ai-skill-scanner (behavioral, trigger, and LLM analysis) and produces SECURITY.md. scan_pr_skills.py scans only changed skills in PRs.

02

Architecture

claude-scientific-skills — Architecture

Distribution

Standalone repository with agentskills.io standard-compatible SKILL.md files. Install via npx skills or manually copy skills.

Install Methods

# Via agentskills CLI (for Cursor/Claude Code/Codex)
npx skills add https://github.com/K-Dense-AI/claude-scientific-skills --skill rdkit
npx skills add https://github.com/K-Dense-AI/claude-scientific-skills --skill scanpy

# Or copy SKILL.md files directly to .claude/skills/ or .cursor/rules/

Directory Tree (top-level)

.
├── scientific-skills/
│   ├── adaptyv/SKILL.md
│   ├── aeon/SKILL.md
│   ├── anndata/SKILL.md
│   ├── ... (138 skill folders)
│   └── zarr-python/SKILL.md
├── docs/                          # Documentation
├── pyproject.toml                 # Package: scientific-agent-skills 2.39.0
├── scan_skills.py                 # Security scanner (all skills)
├── scan_pr_skills.py              # Security scanner (PR-changed skills)
├── README.md
└── SECURITY.md                    # Security scan output

Package

  • Name: scientific-agent-skills
  • Version: 2.39.0
  • Python: >=3.13
  • Runtime deps: cisco-ai-skill-scanner, firecrawl-py, python-dotenv

Each Skill Structure

scientific-skills/<skill-name>/
├── SKILL.md               # Core skill definition
├── references/            # Additional reference docs (in many skills)
└── scripts/               # Example scripts (in some skills)

Target AI Tools

  • Claude Code (primary)
  • Cursor
  • Codex
  • Any agentskills.io-compatible agent
03

Components

claude-scientific-skills — Components

Skills (138)

Organized by domain category:

Bioinformatics & Genomics

anndata, arboreto, biopython, bioservices, bids, cellxgene-census, deeptools, esm, etetoolkit, geniml, gget, gtars, lamindb, phylogenetics, polars-bio, pysam, pydeseq2, scanpy, scikit-bio, scvelo, scvi-tools, tiledbvcf

Cheminformatics & Drug Discovery

datamol, deepchem, diffdock, glycoengineering, matchms, medchem, molfeat, pytdc, rdkit, torchdrug

Healthcare & Clinical

clinical-decision-support, clinical-reports, pyhealth, treatment-plans

Medical Imaging

histolab, pathml, pydicom, pacsomatic

Machine Learning & AI

pytorch-lightning, scikit-learn, scikit-survival, shap, stable-baselines3, torch-geometric, transformers, umap-learn

Scientific Computing

astropy, cobrapy, fluidsim, molecular-dynamics, neurokit2, opentrons-integration, pymatgen, pymc, pymoo, pyopenms, qiskit, qutip, rowan, simpy, statsmodels, sympy

Data Analysis

aeon, dask, exploratory-data-analysis, matplotlib, networkx, polars, seaborn, statistical-analysis, timesfm-forecasting, vaex, zarr-python

Geospatial

geopandas, geomaster

Lab Automation

adaptyv, benchling-integration, dnanexus-integration, ginkgo-cloud-lab, labarchive-integration, omero-integration, protocolsio-integration, pylabrobot

Scientific Communication

bgpt-paper-search, citation-management, docx, exa-search, infographics, latex-posters, literature-review, markdown-mermaid-writing, paper-lookup, paperzilla, pdf, peer-review, pptx, pptx-posters, research-grants, research-lookup, scholar-evaluation, scientific-brainstorming, scientific-critical-thinking, scientific-schematics, scientific-slides, scientific-visualization, scientific-writing, venue-templates, what-if-oracle, xlsx

Database Access

database-lookup (78+ databases), depmap, hugging-science, imaging-data-commons, primekg, usfiscaldata

Cloud & Compute

latchbio-integration, modal, optimize-for-gpu, parallel-web

Miscellaneous

autoskill, consciousness-council, dhdna-profiler, flowio, generate-image, hypogenic, iso-13485-certification, markitdown, market-research-reports, matlab, neuropixels-analysis, open-notebook, pennylane, pufferlib, scikit-learn, stable-baselines3

Security Scripts

  • scan_skills.py — scans all 138 skills using cisco-ai-skill-scanner (behavioral + trigger + LLM analysis)
  • scan_pr_skills.py — PR-scoped scanner for CI integration

No Hooks, No Commands, No Agents

Pure skill collection. No hooks, no subagents, no slash commands.

05

Prompts

claude-scientific-skills — Prompt Excerpts

Excerpt 1: scanpy SKILL.md — Trigger description + structured overview

Technique: Keyword-based auto-trigger description; "When to Use" section prevents wrong-skill loading.

---
name: scanpy
description: Standard single-cell RNA-seq analysis pipeline. Use for QC, normalization, dimensionality reduction (PCA/UMAP/t-SNE), clustering, differential expression, and visualization. Best for exploratory scRNA-seq analysis with established workflows. For deep learning models use scvi-tools; for data format questions use anndata.
license: BSD-3-Clause
metadata:
    skill-author: K-Dense Inc.
---

## When to Use This Skill

This skill should be used when:
- Analyzing single-cell RNA-seq data (.h5ad, 10X, CSV formats)
- Performing quality control on scRNA-seq datasets
- Creating UMAP, t-SNE, or PCA visualizations
- Identifying cell clusters and finding marker genes

Excerpt 2: rdkit SKILL.md — Disambiguation from related skill

Technique: "Use X for Y, use this skill for Z" — active disambiguation prevents wrong-skill selection.

---
name: rdkit
description: Cheminformatics toolkit for fine-grained molecular control. SMILES/SDF parsing, descriptors (MW, LogP, TPSA), fingerprints, substructure search, 2D/3D generation, similarity, reactions. For standard workflows with simpler interface, use datamol (wrapper around RDKit). Use rdkit for advanced control, custom sanitization, specialized algorithms.
---

Excerpt 3: database-lookup SKILL.md — Unified database router

Technique: Single skill routes to 78+ databases by name, avoiding 78 separate skill files.

The database-lookup skill provides access to:

  • PubChem (chemical structures and bioassays)
  • ChEMBL (bioactive molecules)
  • UniProt (protein sequences)
  • COSMIC (cancer mutations)
  • ClinicalTrials.gov (clinical studies)
  • FRED (economic data)
  • USPTO (patents)
  • ...and 71+ more databases
09

Uniqueness

claude-scientific-skills — Uniqueness

Differs from Seeds

No seed comes close in breadth — superpowers has 14 skills, claude-flow has 107, but neither covers scientific domains. The closest archetype is superpowers (skills-only, no commands) but at 10x scale and with domain expertise that superpowers explicitly defers to practitioners. Two features are novel in the entire batch: (1) the security scanner (scan_skills.py using cisco-ai-skill-scanner with behavioral, trigger, and LLM analysis) — no seed or batch peer audits its own skills for adversarial content; (2) the agentskills.io open standard adoption — making the skills portable across Cursor, Codex, and Claude Code without modification.

Positioning

The de facto library for scientific Python workflows with AI coding agents. The 26,000+ stars (highest in the entire batch by 43x) reflect the gap it fills: most AI coding tools are designed for web/app developers, and scientific programmers using RDKit, Scanpy, or OpenMM had no curated skill resources before this.

Observable Failure Modes

  1. Version drift: Scientific Python packages have fast API churn; skill files may document deprecated APIs.
  2. Scale → maintenance debt: 138 skills is a large surface to keep current; the security scanner helps but cannot detect factual drift.
  3. No TDD or validation: Skills document "what to do" but do not enforce testing or output validation.
  4. Context budget: Loading many skills simultaneously (e.g., a multi-omics workflow requiring scanpy + scvi-tools + anndata + networkx) may exhaust context windows.

Explicit Antipatterns

None explicitly documented — the skills are additive rather than prescriptive.

04

Workflow

claude-scientific-skills — Workflow

No Defined Workflow

This is a skill library, not a workflow framework. There are no phases, no approval gates, and no orchestration patterns enforced by the skills themselves.

Skill Activation Pattern

Each skill has a description field with trigger keywords — the agent auto-invokes the skill when the user's request matches the keywords. For example:

  • scanpy — triggers on "single-cell RNA-seq", "QC", "UMAP", "clustering"
  • rdkit — triggers on "SMILES", "molecular descriptors", "fingerprints", "cheminformatics"
  • database-lookup — triggers on database names (PubChem, ChEMBL, UniProt, etc.)

Skill Structure Pattern

Each skill follows a consistent structure:

  1. Overview (when to use)
  2. Quick start (imports, setup)
  3. Core capabilities (organized by use case)
  4. Common patterns (code examples)
  5. Best practices
  6. Related skills (cross-references)

Security Scan Workflow (CI)

# Run by maintainers before release
python scan_skills.py  # → SECURITY.md
python scan_pr_skills.py  # In PRs, scans only changed skills
06

Memory Context

claude-scientific-skills — Memory & Context

State Storage

None. Pure skill collection with no persistent state.

Context Injection

Each skill's SKILL.md is loaded on-demand when trigger keywords match. Many skills include references/ folders with additional documentation:

  • scanpy/references/ — extended analysis patterns
  • rdkit/references/ — molecular manipulation guides
  • pytorch-lightning/references/ — training patterns

Cross-Session Handoff

None.

Progressive Context Loading

Skills use cross-references to other skills: "For deep learning models use scvi-tools; for data format questions use anndata." This creates a navigable skill graph rather than a monolithic context dump.

Security Context

SECURITY.md documents the results of periodic security scans — providing transparency about behavioral analysis findings for all 138 skills.

07

Orchestration

claude-scientific-skills — Orchestration

Multi-Agent Pattern

None. Skills are invoked individually; no orchestration.

Execution Mode

Interactive-loop — skill is loaded on user request trigger.

Isolation Mechanism

None.

Multi-Model

No. Agent uses session default model.

Auto-Validators

None.

Prompt Chaining

None defined. Users can chain skills manually.

Multi-Agent Capability

The parallel-web skill provides guidance on parallel web scraping — this is not a multi-agent orchestration framework but a skill for concurrent HTTP requests.

08

Ui Cli Surface

claude-scientific-skills — UI / CLI Surface

CLI Binary

None from this repo. The K-Dense BYOK companion app provides a UI, but it is a separate product.

Local UI

None.

Install CLI

npx skills add https://github.com/K-Dense-AI/claude-scientific-skills --skill <name>

The skills CLI is an external tool from the agentskills.io ecosystem, not bundled in this repo.

IDE Integration

Compatible with:

  • Claude Code
  • Cursor (via .cursor/rules/)
  • Codex

Observability

SECURITY.md — generated by scan_skills.py, provides security scan results for all 138 skills.

Related frameworks

same archetype · same primary tool · same memory type

OpenHarness ★ 13k
A11 Governance

Open-source Python agent runtime providing complete harness infrastructure: tools, memory, governance, swarm coordination, and…

Trae Agent ★ 12k
A11 Governance

Research-friendly open-source CLI coding agent by ByteDance, designed for academic ablation studies and modular LLM provider…

Sweep AI ★ 7.7k
A11 Governance

Autonomous GitHub bot that converts issues to pull requests using a sequential multi-agent pipeline.

Agent Governance Toolkit (microsoft) ★ 2.3k
A11 Governance

Enterprise-grade AI agent governance: YAML policy enforcement, 12-vector prompt injection defense, zero-trust identity,…

TDD Guard ★ 2.1k
A11 Governance

Mechanically enforces the Red-Green-Refactor TDD cycle by blocking file writes that violate TDD principles via a PreToolUse hook…

Agentic Coding Flywheel Setup (ACFS) ★ 1.5k
A11 Governance

Take a complete beginner from laptop to three AI coding agents running on a VPS in 30 minutes via an idempotent manifest-driven…