A curated pack of scientific skills for AI coding agents — Claude Code, Codex, Cursor, OpenCode, Copilot, Windsurf, Gemini, and ~45 others via the skills.sh ecosystem. Each skill is a self-contained capability bundle covering a specific computational-biology, genomics, statistics, or scientific-reporting workflow — frontmatter, prose guidance, reference documents, and ready-to-run scripts.
The agent reads each skill's SKILL.md, follows the documented protocol, and runs the bundled scripts directly. Install once, use from whichever agent you have configured.
Skills are grouped by topic. Each lives at skills/<skill-name>/ with its own SKILL.md, optional references/, and optional scripts/.
| Topic | Skills |
|---|---|
| Single-cell transcriptomics | scrnaseq-scanpy-core-analysis · scrnaseq-seurat-core-analysis · scrna-trajectory-inference · scrna-disease-drug-discovery · cell-cell-communication · spatial-transcriptomics |
| Bulk transcriptomics | bulk-rnaseq-counts-to-de-deseq2 · bulk-rnaseq-differential-expression · bulk-omics-clustering · functional-enrichment-from-degs |
| Epigenomics (ChIP-seq) | chip-atlas-diff-analysis · chip-atlas-peak-enrichment · chip-atlas-target-genes |
| Statistical genetics | genetic-variant-annotation · gwas-to-function-twas · mendelian-randomization-twosamplemr · polygenic-risk-score-prs-catalog · genetic-target-hypothesis |
| Functional genomics & networks | pooled-crispr-screens · gene-essentiality · coexpression-network · grn-pyscenic · upstream-regulator-analysis |
| Proteomics & multi-omics | proteomics-diff-exp · multi-omics-integration · pathway-enrichment |
| Clinical & translational | survival-analysis-clinical · disease-progression-longitudinal · clinicaltrials-landscape · lasso-biomarker-panel |
| Wet-lab & external APIs | pcr-primer-design · adaptyv-api |
| Scientific reporting | docx-generation · pptx-generation · pdf-report-generation |
| Methodology & meta-skills | experimental-design-statistics · data-analysis-best-practices · literature-preclinical · skill-creator |
The pack is published in the skills.sh open-agent ecosystem. The same command installs into whichever agent you have configured (Claude Code, Codex, Cursor, OpenCode, Copilot, Windsurf, Gemini, …):
# Install the whole pack
npx skills add 001TMF/agentic-science
# Install at the user level (~/<agent>/skills/) so every project can use it
npx skills add 001TMF/agentic-science -g
# Install only specific skills
npx skills add 001TMF/agentic-science --skill scrnaseq-scanpy-core-analysis --skill bulk-rnaseq-counts-to-de-deseq2
# Pin to a specific agent
npx skills add 001TMF/agentic-science -a claude-code
# Just list what's available
npx skills add 001TMF/agentic-science --listnpx skills handles per-agent path conventions automatically (.claude/skills/, .cursor/rules/, .codex/skills/, etc.) and supports both --copy and symlinked installs. See skills.sh/docs for the full option set.
If you don't have Node available, the repo ships a small bash installer that targets Claude Code's .claude/skills/ path:
git clone https://github.com/001TMF/agentic-science.git
cd agentic-science
./install.sh # into ./.claude/skills/
./install.sh ~/my-project # into a specific project
./install.sh --global # into ~/.claude/skills/
./install.sh --only scrnaseq-scanpy-core-analysis,bulk-rnaseq-counts-to-de-deseq2
./install.sh --list # list without installingThe pack is just files. Copy whatever you need:
cp -r skills/scrnaseq-scanpy-core-analysis /path/to/your-project/.claude/skills/Or copy everything:
cp -r skills/* /path/to/your-project/.claude/skills/Open your agent in the project and call a skill by name:
> Use the scrnaseq-scanpy-core-analysis skill to QC and cluster this dataset.
The agent reads the skill's SKILL.md, follows the documented workflow, and runs the bundled scripts directly from scripts/.
Every skill follows the same shape so they remain interchangeable:
skills/<name>/
├── SKILL.md # Frontmatter + canonical sections (When-to-Use, Inputs,
│ # Outputs, Clarification Questions, Workflow, Common
│ # Issues, Suggested Next Steps, References)
├── references/ # Optional. Methodology, troubleshooting, parameter
│ ├── *.md # tables, deeper guidance the agent can pull in.
├── scripts/ # Optional. Runnable Python / R / shell scripts that
│ ├── *.py # implement the workflow steps.
│ ├── *.sh
│ └── *.R
For the full authoring standard — frontmatter schema, the canonical 14 sections, when to add references, when to add scripts, the 7-point script style guide — see the companion guide repo at https://github.com/001TMF/writing-scientific-agent-skills.
Each skill meets these gates:
- YAML frontmatter with
id,name,display-name,short-description,category,keywords,version,last-updated. - All 7 mandatory canonical sections (When-to-Use with ❌ anti-patterns, Inputs, Outputs, Clarification Questions with
⚠️ CRITICAL: ASK THIS FIRSTon Q1, Common Issues table with ≥10 rows, Suggested Next Steps, References). - Scripts ship with module docstrings, type hints, argparse
--help, success verification prints, gracefulImportError,__main__guard, no hardcoded paths. - References use canonical terminology and cite primary sources.
- License-compliance table for any third-party packages (commercial use clearly marked).
MIT. See LICENSE.
Open an issue or PR. New skills should follow the structure documented at https://github.com/001TMF/writing-scientific-agent-skills.