Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add zamushwani/biomedical-ai-skills --skill foundation-modelsgit clone --depth 1 https://github.com/zamushwani/biomedical-ai-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/zamushwani/biomedical-ai-skills/foundation-models)<a href="https://agentmods.dev/skills/zamushwani/biomedical-ai-skills/foundation-models"><img src="https://agentmods.dev/badge/skills/zamushwani/biomedical-ai-skills/foundation-models/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/zamushwani/biomedical-ai-skills/foundation-models"><img src="https://agentmods.dev/badge/skills/zamushwani/biomedical-ai-skills/foundation-models.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00000 | $0.04174 |
| Opus 5 | $0.00000 | $0.02087 |
| Sonnet 5 | $0.00000 | $0.00835 |
| Haiku 4.5 | $0.00000 | $0.00417 |
Grade A, and why
foundation-models scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 361 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Single-Cell Foundation Models
When to use scGPT, Geneformer, UCE, and the perturbation models, and when a linear baseline beats them. Covers zero-shot embedding extraction, fine-tuning for annotation, in-silico perturbation, and the benchmark evidence behind each recommendation.
When to Use This Skill
Activate when the user requests:
- scGPT or Geneformer embeddings, fine-tuning, or in-silico perturbation
- Zero-shot cell type annotation with a pretrained transformer
- Cross-species or reference-free cell embedding (UCE, TranscriptFormer)
- Perturbation effect prediction (STATE, Tahoe-x1, GEARS)
- Advice on whether a foundation model is worth it for a given task
The Short Answer
Default: do not use one.
For human cell type annotation, CellTypist with a matching pretrained model, or scANVI/scVI for joint integration and label transfer, gives an answer in minutes on a CPU or small GPU, with a failure mode you can inspect. Three independent 2026 benchmarks put scVI or plain PCA at or above every foundation model on integration and representation.
The evidence base has converged: single-cell foundation models are representation-strong and prediction-weak.
Where they reliably lose to simple baselines:
Zero-shot clustering and annotation vs HVG + PCA, scVI
Batch integration vs scVI, Harmony
Trajectory inference from embeddings vs HVG (temporal compression)
Gene expression reconstruction vs predicting the mean
Single-gene perturbation prediction vs additive / linear models
GRN inference from attention weights vs trivial co-expression baselines
That is not a reason to never use one. It is a reason to know which situation you are in.
When to Reach for One
| Situation | Model | Why |
|---|---|---|
| <500-1000 labeled cells, large unlabeled target | Fine-tuned scGPT whole-human or Geneformer V2-104M | The one setting where the pretraining advantage is documented to grow as labels shrink |
| Query cell types absent from any CellTypist/Azimuth reference | scFM embedding + clustering, or UCE zero-shot | Reference-free by construction |
| Non-human or multi-species | TranscriptFormer or UCE | Shared cross-species latent space |
| Spatial and dissociated data jointly | Nicheformer | The only mature spatial foundation model |
| Combinatorial perturbations, cross-context transfer | Arc STATE or Tahoe-x1 | The one place deep models beat linear |
| Cancer / drug response | Tahoe-x1, Geneformer-V2-104M_CLcancer, or scGPT pan-cancer | Domain-matched pretraining |
| Bulk RNA-seq to prognosis | scFoundation embeddings as extra features | Documented C-index gains, low redundancy with expression |
What ships with it
6 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 361 lines · 0 tokens per session scan A 4268f6fc635e
foundation-models is a skill published in the GitHub repository zamushwani/biomedical-ai-skills (1 stars, last pushed 13d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 4,174 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
cerna-analysis
Use when building a ceRNA regulatory network from a key gene list by combining bundled miRNA-mRNA and miRNA-lncRNA database files, with flat-file CSV exports and PDF visualization in a single output directory. NOT for: differential expression, single-cell analysis, enrichment analysis, or workflows without a key gene…
cibersort-immune-infiltration-analysis
Use when estimating relative immune cell infiltration from a bulk expression matrix with a CIBERSORT-style nu-SVR deconvolution workflow based on an LM22 signature matrix, comparing one case group against one control group, and generating structured tables plus immune-fraction plots. NOT for single-cell RNA-seq…
gene-protein-expression-matrix-normalization
Use when normalizing bulk gene or protein expression matrices with log2 transform, z-score standardization, or min-max scaling before downstream visualization or exploratory analysis. NOT for count-model normalization such as TPM/DESeq2 size factors, batch correction, or single-cell preprocessing.
genomic-intelligence
Predict regulatory features, gene structure, and expression directly from DNA sequence using Genomic Intelligence's hosted transformer DNA language models — no local GPU or model weights. Six tasks over a REST API and a hosted MCP server (keyless public demo): promoter regions, splice donor/acceptor sites, enhancer…
torch-geometric
PyTorch Geometric (PyG) for graph neural networks — node/link/graph classification, message passing (GCN, GAT, GraphSAGE, GIN), heterogeneous graphs, neighbor sampling, and custom datasets. Use when working with torchgeometric, not for general NetworkX analytics or non-graph PyTorch models.
bids
Use this skill when working with Brain Imaging Data Structure (BIDS) datasets: organizing neuroscience and biomedical data (MRI, EEG, MEG, iEEG, PET, microscopy, NIRS, motion capture, EMG, MR spectroscopy, behavioral), querying BIDS layouts, validating compliance, converting DICOM to BIDS, writing metadata sidecars…