Scientific Agent Skills
K-Dense-AI/scientific-agent-skills
177 research skills for databases, Python science packages, lab platforms and scientific writing.
Scientific Agent Skills is an MIT-licensed library of 177 skills from K-Dense Inc. that gives an agent procedures for scientific work: querying public databases, using packages such as Scanpy, RDKit and Astropy, running lab and cloud platforms, and writing papers. It follows the Agent Skills standard and is also packaged as an Agent Plugins bundle. The authors published a paper describing the library.
Facts checked on against the sources listed at the end of this page.
- GitHub stars
- 47,399
- Skills in the pack
- 177
- Agents supported
- 8Plugin install is client-specific; check your host's current documentation
- Instruction size
- 311k wordsacross all SKILL.md files
How to install Scientific Agent Skills
| Agent | Command | Copy |
|---|---|---|
| Any agent | Installs into the ~/.agents/skills/ convention, or .agents/skills/ for a project. Confirm the path against your host. | |
| Claude Code | Needs GitHub CLI 2.90.0 or later. Add a skill name to install just one, or --pin to fix a version. | |
| Codex | The README also lists codex plugins install . from a clone, using the root plugin.json. | |
| Cursor | Or symlink a clone into ~/.cursor/plugins/local and reload the window. | |
| Gemini CLI | |
Commands as given in the README on 2 October 2026.
What is inside Scientific Agent Skills
| Skill | What it makes the agent do |
|---|---|
| database-lookup | One skill covering 80 public, registered or licensed databases such as PubChem, UniProt, ClinVar, FRED and SEC EDGAR, with provenance for each result. |
| paper-lookup | Searches PubMed, PMC, bioRxiv, arXiv, OpenAlex, Crossref, Semantic Scholar and other literature sources. |
| scientific-writing | Evidence-traceable drafting of papers, with citation management and venue templates alongside. |
| statistical-analysis | Statistical workflows, with companion skills for power analysis, experimental design and measurement uncertainty. |
| geopandas | Geospatial analysis, paired with GeoMaster for remote sensing and spatial machine learning. |
| cantera | Chemical kinetics: ignition delay, mechanism snapshots, numerical refinement and conservation checks. |
| pybamm | Battery experiments: charge and discharge protocols, parameter provenance and comparison with measured curves. |
| nextflow | Builds, runs and debugs Nextflow and nf-core pipelines. |
| optimize-for-gpu | Rewrites CPU-bound Python to use CuPy, Numba CUDA, cuDF, cuML and related libraries. |
| generate-image | Image generation and editing through the OpenRouter Image API, with dry runs. |
Ten of 177 skills are shown. The README groups the rest into categories such as bioinformatics, cheminformatics, clinical, protein engineering, lab automation, materials science and regulatory standards.
What Scientific Agent Skills runs on your machine
| Hooks | None named in the README. |
|---|---|
| Bundled scripts | Yes. Many skills ship Python scripts, and the repo keeps a test suite for each one. Dependencies are installed per skill, with uv and Python 3.13 or later listed as prerequisites. |
| Network calls | Yes. Database, literature, hosted-model and cloud-platform skills call external services. Local analysis can work offline once packages and reference data are present. |
| API keys needed | Some. Skills for registered or paid services, such as OpenRouter image generation, need their own API keys. Check each SKILL.md. |
| Tool permissions | The README has a security disclaimer. It advises reading each full SKILL.md and its scripts, scanning before install and pinning a version. Every skill is scanned with Cisco's Skill Scanner, which the README says does not guarantee safety. |
These are facts read from the repo, not a security rating. Read the files before you install any skill.
How much context Scientific Agent Skills uses
| All 177 skills | 310,854 words across every SKILL.md file |
|---|---|
| Largest skill | imaging-data-commons, 4,071 words |
| Loaded at start | Only each skill's name and description. The full text loads when a task needs that skill. |
Word counts are measured from the repo every day. Reference files a skill loads on demand are not counted.
How Scientific Agent Skills works
The library covers a lot of ground: genomics and single-cell analysis, cheminformatics, clinical data, protein design, physics and engineering simulation, lab automation and scientific communication. Most skills wrap one package, database or platform and tell the agent how to use it correctly.
The README says each skill carries its own license field in its SKILL.md, which can differ from the repo's MIT license. K-Dense also sells a hosted platform, and the README describes when each fits. The skills themselves are free.
A good fit for
- Researchers who want an agent to use specialist Python packages and public databases correctly
- Labs that need literature search, analysis and paper writing in one install
- Teams that want to pin a version and install only the skills they use
- Bioinformatics and cheminformatics work, which make up much of the library
Skip it if
- General software development, which the repo says is outside its scope
- Anyone who wants a small install, since the full set is 177 skills with many dependencies
- Clinical, regulatory or compliance decisions, since the skills say they do not replace qualified review
Is Scientific Agent Skills still maintained?
| Last commit | , 1 day ago |
|---|---|
| Latest release | v2.72.0, 1 October 2026 |
| Stars gained in 7 days | Shown after 7 days of tracking |
| Stars gained in 30 days | Shown after 30 days of tracking |
| Open issues and pull requests | 21 |
| Forks | 4,286 |
Tools Scientific Agent Skills works with
Questions about Scientific Agent Skills
Is Scientific Agent Skills free?
Yes. The repo is MIT licensed. The README notes that each skill has its own license field, so check the ones you use.
Do I need to install all 177 skills?
No. gh skill install accepts a single skill name, and you only need the Python packages for the workflows you run.
Does it work offline?
Local analysis can, once dependencies, reference data and model files are in place. Database queries, hosted models and cloud platforms need a network connection.
Does it work with agents other than Claude Code?
Yes. It follows the open Agent Skills standard and installs for Cursor, Codex, Gemini CLI and other hosts. It is also an Agent Plugins package.
Is it safe to install?
The authors scan every skill with Cisco's Skill Scanner and review contributions by hand, but they say a clean scan is not a guarantee. They advise reading SKILL.md files and scripts before installing and pinning a version.