Nanosaur2-670M
well9472 · Kit
Makes illustrations from text prompts with a small 670M-parameter diffusion model, run through ComfyUI. Built as a low-budget research model for people who wan…

K-Dense’s open library of more than 170 research skills for coding agents: biology, chemistry, medicine, materials, data analysis, literature search and scientific writing. Install the few you need, and read each skill’s license and needs first.
Scientific Agent Skills is an open-source project on GitHub in the Research category, listed by Mindket. It is free. This is version 2, last updated on . Install it from GitHub as its README says. Get Mindket’s setup notes on your Mindket account, as a zip.
npx skills add K-Dense-AI/scientific-agent-skills
And 2 more in Where to use it
Scientific Agent Skills is K-Dense’s library of research skills for coding agents, with more than 170 skill folders. Each one teaches an agent a package, a database or a job. Examples are Scanpy for single-cell data, RDKit for molecules, a lookup across 18 scholarly APIs and manuscript drafting with traceable evidence. The agent can use any Python package without a skill, and these add curated notes, examples and checks. Install only the ones you need.
People who do computational research with an AI agent and want it to follow the conventions of a field. That includes biologists, chemists, data analysts and anyone writing or reviewing papers. It is not a replacement for expertise. Its clinical skills are for research and drafts that qualified people review, never for diagnosis or treatment choices.
| Who | Skills they reach for | Example |
|---|---|---|
| Biologists and bioinformaticians | scanpy, pydeseq2, database-lookup | A researcher runs quality control, clustering and marker ranking on single-cell counts that come with known provenance. |
| Chemists and drug research teams | rdkit, datamol, diffdock | A chemist parses molecules from SMILES and computes descriptors and fingerprints, or searches for substructures. |
| Authors and reviewers | scientific-writing, literature-review, peer-review | An author drafts a manuscript with each claim tied to a verified source, and a reviewer prepares notes on a manuscript they are authorized to read. |
| Data analysts | statistical-analysis, scientific-visualization | An analyst picks a test, checks its assumptions, reports effect sizes and builds a figure. |
Any host that reads Agent Skills. The README names Cursor, Claude Code, Codex, Google Antigravity and Gemini CLI, and lists macOS, Linux or Windows with WSL2 as the systems.
Get it saves a zip of Mindket’s setup notes in your downloads folder, with a route for each agent the README names and notes on Python and keys.
| Need | Detail |
|---|---|
| An agent | Claude Code, Codex, Cursor, Gemini CLI, Antigravity or another host that reads Agent Skills |
| uv | The README says this Python package manager is required for installing skill dependencies |
| Python | 3.13 or newer for the repository’s own tooling, and each skill states its own range |
| A system | macOS, Linux or Windows with WSL2 |
| An installer | Node.js 22.20 or newer for npx, from the skills installer’s README, or GitHub CLI 2.90.0 or newer for gh skill |
| Keys | They depend on the skill, and many need none |
npx skills add K-Dense-AI/scientific-agent-skills
The GitHub CLI route installs one named skill directly. The README says it needs GitHub CLI 2.90.0 or newer:
gh skill install K-Dense-AI/scientific-agent-skills scanpy
A skill’s packages come next, and uv installs them. Scanpy’s own SKILL.md gives this line and asks for Python 3.12 or newer:
uv pip install "scanpy[leiden]"
uv’s own docs say uv pip install needs a virtual environment by default. So make one first in your work folder with uv venv, adding --python 3.12 for Scanpy. That step is Mindket’s note.
Describe the whole job, and the agent picks the skills that match. The README’s example prompts open with a line asking the agent to use available skills whenever possible.
| Skill | What you give it | What comes back |
|---|---|---|
| database-lookup | A fact to fetch, with a named source and filters | Results from the databases that fit, among the 80 it documents, with endpoints, pagination and provenance |
| paper-lookup | A topic, DOI, PMID, arXiv id or author | Papers, citations and open-access links, with the endpoints, parameters and access date |
| literature-review | A review question | A search record, screening counts and a Markdown or PDF review with checked citations |
| scanpy | Single-cell count data with known provenance | Quality control, clustering, marker ranking and plots |
| scientific-writing | The evidence you hold | Manuscript text with claims tied to verified sources and a human check |
| statistical-analysis | A comparison or a hypothesis | A test choice, assumption checks, effect sizes and a report in APA style |
You need open-access papers on base editing for a background section, and you want to repeat the search later.
Use available skills you have access to whenever possible. Find open-access papers on CRISPR base editing for sickle cell disease since 2023, and give me each DOI with the database it came from.
Claude reads the paper-lookup skill, queries the databases that fit the request, and answers with titles and DOIs plus the endpoints, parameters and access date, so you can repeat the search.
If you are new. A skill is a folder of instructions that your agent reads when your request matches it. Start with one skill for a job you already do by hand, such as a literature search. Read its SKILL.md first. It says what the skill connects to and what must be installed. Leave the rest uninstalled.
For experts. Most SKILL.md headers hold a license, a compatibility line and a version. The README says skills with bundled scripts have a test suite under tests/<skill-name>/ and an entry in tests/skill-requirements.toml. Pin a release for repeatable installs, and keep one environment per workflow, since package pins can clash. The README’s lines for pinning and updating follow. Swap in the release tag you want:
gh skill install K-Dense-AI/scientific-agent-skills --pin v2.71.0 gh skill update --all
Some skills, such as database-lookup and paper-lookup, tell the agent to treat what a database returns as untrusted data. The database-lookup skill also has the agent ask you before a retrieval over 10,000 records or 100 API calls.
It is a large collection, so start small.
# Scientific Agent Skills: setup notes This kit holds setup notes written by Mindket. It does not contain the skills, their scripts or their reference files. Those install from K-Dense’s repository: https://github.com/K-Dense-AI/scientific-agent-skills License: MIT, Copyright (c) 2025 K-Dense Inc. (the license file’s text). Full text: https://github.com/K-Dense-AI/scientific-agent-skills/blob/main/LICENSE.md The repository’s own README adds that each skill has its own license in the license line of its SKILL.md, and that these can differ from the repository’s MIT license and carry extra terms. Mindket read the first 40 lines of every SKILL.md: most say MIT, BSD or Apache, and a few say something else, such as a noncommercial license (PolyForm Noncommercial for deepspot-m, CC BY-NC-SA 4.0 for what-if-oracle), GPL (bioservices, cobrapy, etetoolkit) or a proprietary service (rowan). Three say Unknown and four have no license line at all. Read the license line of each skill you install. The kit holds none of those skills, so it carries none of their terms. Scientific Agent Skills is a community project by K-Dense. It is not made or endorsed by Anthropic. The pages these notes came from: the README (https://github.com/K-Dense-AI/scientific-agent-skills/blob/main/README.md), the repository’s plugin.json and SECURITY.md, and the SKILL.md files of the skills named in these notes, under https://github.com/K-Dense-AI/scientific-agent-skills/tree/main/skills. The GitHub CLI lines come from GitHub’s manual for gh skill install and gh skill update (https://cli.github.com/manual/gh_skill_install). Facts about where Claude Code loads skills come from Claude Code’s own skills page (https://code.claude.com/docs/en/skills). Mindket copied every command and ran none of them. ## What it is A library of research skills for coding agents: more than 170 folders under skills/, each with a SKILL.md and, for many, reference files, scripts and templates. The README groups them by field: bioinformatics, drug discovery, proteomics, clinical research workflows, microscopy and medical imaging, neuroscience, machine learning, materials, engineering, geospatial work, laboratory automation, scientific writing and review, and database lookup across 80 sources. The README says the agent can use any Python package on its own. A skill adds curated notes, examples and checks for one package, database or job, so the agent starts from written guidance. The README adds that community skills may have been reviewed less closely than K-Dense’s own. ## Who does what - You: install only the skills you need, see that each skill’s Python packages are installed in their own environment (you, or the agent following the SKILL.md’s install lines), give any API key that skill asks for, and check what the agent produces. - The installer, either npx skills or gh skill: puts skill folders where your agent reads them. npx skills can link them to one shared copy or copy them. It does not install what a skill needs to run. - The agent (Claude, Codex or another): reads a skill when your request matches it, then runs the commands and scripts the skill names. The skills’ imperatives are written to the agent. - The skills’ scripts: run on your machine, and some call web services. The README says skills can run code, install packages, make network requests and change files. ## Files in this kit - setup/choose-skills.md: the README’s security advice, how to list and pick a few skills, and how to read each skill’s license and needs - setup/claude-code.md: the npx and gh skill routes for Claude Code - setup/codex.md: the same routes for Codex, and the README’s plugin line - setup/cursor.md: the same routes for Cursor, and the README’s plugin symlink route - setup/other-agents.md: Gemini CLI, Antigravity, and agents that scan ~/.agents/skills, with the README’s clone lines - setup/python-and-keys.md: uv, separate environments, the system the README names, and example API keys ## What you need first - An agent that reads Agent Skills. The README names Cursor, Claude Code, Codex, Google Antigravity, Gemini CLI and Claude Cowork. - uv, a Python package manager. The README says it is required for installing skill dependencies. - Python. The README asks for 3.13 or newer for the repository’s own tooling, and each skill states its own range. - macOS, Linux, or Windows with WSL2. That is the README’s list. - Node.js for the npx route (the skills installer’s README asks for 22.20 or newer), or the GitHub CLI version 2.90.0 or newer for the gh skill route. - Whatever each skill says it needs. Some skills need no key, some take an optional one, and some need an account with a service. ## What installs where - npx skills add: the skill folders, in the agent’s own skills folder, in the project or globally. - gh skill install: the same, with a note about the source added to each SKILL.md’s header, which lets gh skill update find changes. - Plugin route, for Cursor and Codex: the repository as one plugin, with every skill under skills/. - Python packages, system tools and keys: nowhere. Each skill’s SKILL.md says what it needs, and installing the skill files installs none of it. ## Your data The skill files are text and scripts. What a skill does depends on the skill: some work only on files on your computer, and others call public databases or services such as PubMed, ClinVar or OpenRouter. The README’s own advice is to read each SKILL.md first, since it says which external services the skill connects to. What you give your agent goes to your agent’s provider as in any other work. ## Update and remove gh skill update (and its --all option) is in setup/claude-code.md. npx skills update and npx skills remove are in the skills installer’s README (https://github.com/vercel-labs/skills). The maker’s README gives no removal step. Claude Code’s skills page says a personal or project skill is removed by deleting its folder.
README.md 6 KB setup/choose-skills.md 5 KB setup/claude-code.md 4 KB setup/codex.md 2 KB setup/cursor.md 2 KB setup/other-agents.md 3 KB setup/python-and-keys.md 4 KB
Updated by Mindket to the current version of this kit.
Listed GitHub projects in the same category that share the most topics with this one.
| Product | Stars | Activity | Language | License | Last change |
|---|---|---|---|---|---|
| Scientific Agent Skills This page | 56,365 | — | — | MIT | — |
| bioinformatics-agent-skillsShares bioinformatics · claude-skills · computational-biology | 75 | Active | Python | MIT | |
| DYZ MedPaperShares clinical-research · agent-skills | 236 | Active | Python | — |
Posts and comments that name this project on Hacker News, Bluesky, X, Reddit, Facebook and YouTube, found by a weekly search and on Trendshift. Mindket doesn’t write or check them. Open one to read it where it was posted.
well9472 · Kit
Makes illustrations from text prompts with a small 670M-parameter diffusion model, run through ComfyUI. Built as a low-budget research model for people who wan…
Waqf-AI · Kit
Reads handwritten Arabic text from images and returns it as text. A fine-tuned PaddleOCR-VL model for people digitizing Arabic handwriting.
Wan-AI · Kit
Makes video clips from a text prompt or an image plus text, at up to 720P and 24 frames per second, in the Diffusers format for a single consumer GPU.
ssinder · Skill
Gives an AI coding agent expert skills for computational biology, from RNA-seq and single-cell analysis to protein structure, molecular dynamics, peptide desig…
michael-denyer · Skill
Michael Denyer’s port of Lauren Tan’s pstack skill stack to Claude Code, Codex, Pi, GitHub Copilot, Muse and other agents. Tell its poteto-mode skill your goal…
dietrichgebert · Skill
Guides AI coding agents to write minimal code by prioritizing existing repository components and standard libraries. It adds auditing, code review, and trackin…