biomedical-ai-skills
AGENTS.md1 starsChanged 3 months agoMITView on GitHub
Validated protocol files for cancer bioinformatics and multi-omics research. This repository is a library of 16 SKILL.md files. Each is a self-contained protocol for one domain: what to run, which parameter values and why, and the mistakes that produce plausible-looking wrong answers. They are plain Markdown with no tool-specific syntax, so any agent can read them directly. Read the relevant skills/<name>/SKILL.md before writing analysis code in that domain, and follow it in place of a default approach. Each file states…
- Installs packages
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository18 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .cursor/rules/biomedical-skills.mdcCursor rule
- biomedical-mcpSkill
- cancer-multiomicsSkill
- checkpoint-biomarkersSkill
- clinical-nlpSkill
- computational-pathologySkill
- drug-responseSkill
- epigenomicsSkill
- foundation-modelsSkill
- immune-deconvolutionSkill
- meta-analysisSkill
- multiomics-integrationSkill
- radiotherapy-responseSkill
- single-cell-atlasSkill
- spatial-transcriptomicsSkill
- survival-analysisSkill
- variant-annotationSkill
The file
We split this AGENTS.md into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 9
Read the relevant skills/<name>/SKILL.md before writing analysis code in that domain, and follow it in place of a default approach.
- Very specific: line 30
Every skill carries a README.md (a gotchas summary) and a tests/ directory validating its claims against public data or live registries.
- Distinctive: line 36
Verify before you write: Never state a version number, function signature, argument name or default from memory.
- Distinctive: line 39
Never claim a test passes unless you executed it: If it was not run, say "written but not executed" and say why.
- Distinctive: line 40
Stage explicit paths: Never git add . — untracked personal files sit in the repository root.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md
Validated protocol files for cancer bioinformatics and multi-omics research.
This repository is a library of 16 SKILL.md files. Each is a self-contained protocol for one domain: what to run, which parameter values and why, and the mistakes that produce plausible-looking wrong answers. They are plain Markdown with no tool-specific syntax, so any agent can read them directly.
Using a skill
Read the relevant skills/<name>/SKILL.md before writing analysis code in that domain, and follow it in place of a default approach. Each file states when it applies in its own ## When to Use This Skill section.
| Skill | Domain |
cancer-multiomics | Expression, mutation, CNV and methylation for TCGA/GEO |
immune-deconvolution | Immune and stromal composition from bulk RNA-seq |
survival-analysis | Kaplan-Meier, Cox, competing risks, RMST |
single-cell-atlas | scRNA-seq from raw counts to annotation and trajectory |
spatial-transcriptomics | Visium, Xenium, MERSCOPE, CosMx |
foundation-models | scGPT, Geneformer, UCE, and when a linear baseline wins |
meta-analysis | Systematic review, pooling, network meta-analysis |
variant-annotation | VCF normalization, ACMG/AMP, oncogenicity, TMB, MSI |
drug-response | Dose-response fitting, GDSC/CTRP, sensitivity prediction |
clinical-nlp | Clinical text extraction, assertion, de-identification |
computational-pathology | Whole-slide imaging, tiling, MIL, cell segmentation |
biomedical-mcp | MCP servers over GDC, GEO and biomarker databases |
multiomics-integration | MOFA+, SNF, iClusterPlus, DIABLO |
checkpoint-biomarkers | PD-L1, TMB, MSI, IFN-gamma/TIS/TIDE signatures |
radiotherapy-response | DDR profiling, Radiosensitivity Index, GARD |
epigenomics | ATAC-seq and ChIP-seq, peak calling, differential binding |
Every skill carries a README.md (a gotchas summary) and a tests/ directory validating its claims against public data or live registries.
Working in this repository
The rules that govern changes here live in CLAUDE.md and apply to any agent, not only Claude Code. The ones that matter most:
- Verify before you write. Never state a version number, function signature, argument name or default from memory. Check it against the registry and the package source before it goes in a file.
- Changing a skill means releasing it. A committed but unreleased edit leaves
pip installserving stale content. - Never report success from an intermediate signal. A green CI run is not evidence the artifact is correct. Verify the artifact itself.
- Never claim a test passes unless you executed it. If it was not run, say "written but not executed" and say why.
- Stage explicit paths. Never
git add .— untracked personal files sit in the repository root.
Commands
python3 tools/run_benchmarks.py # run every skill's validation suite
python3 tools/check_portability.py # verify the skills stay tool-neutral
cd skills/<name>/tests && python run_all.py # one skill (some use run_all.R)Layout
skills/<name>/SKILL.md the protocol
skills/<name>/README.md gotchas summary
skills/<name>/tests/ validation suite
src/biomedical_ai_skills/ pip package and CLI
tools/ benchmark and portability runnersA note on portability
The SKILL.md files deliberately contain no agent-specific syntax: no argument placeholders, no file-include directives, no inline shell execution. That is what lets the same file work in Claude Code, Codex CLI, Cursor, Gemini CLI and anything else that reads Markdown. tools/check_portability.py enforces it, so a skill that acquires tool-specific syntax fails the check rather than silently breaking elsewhere.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.