ToolUniverse is a collection of tools, interfaces, and supporting components for building AI systems that perform scientific work. It is for developers creating AI scientist agents that use APIs, databases, machine-learning tools, and domain-specific utilities. The catalogue includes skills, commands, an MCP server, an agent, and a hook for working with the ecosystem.
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 mims-harvard/ToolUniverse --skill tooluniverse-scvi-toolsgit clone --depth 1 https://github.com/mims-harvard/ToolUniverseWrote 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/mims-harvard/tooluniverse/tooluniverse-scvi-tools)<a href="https://agentmods.dev/skills/mims-harvard/tooluniverse/tooluniverse-scvi-tools"><img src="https://agentmods.dev/badge/skills/mims-harvard/tooluniverse/tooluniverse-scvi-tools/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/mims-harvard/tooluniverse/tooluniverse-scvi-tools"><img src="https://agentmods.dev/badge/skills/mims-harvard/tooluniverse/tooluniverse-scvi-tools.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 | $0.00204 | $0.01342 |
| Opus 5.5 | $0.00082 | $0.00537 |
| Sonnet 5.5 | $0.00041 | $0.00268 |
| Haiku 4.5 | $0.00020 | $0.00134 |
Grade A, and why
tooluniverse-scvi-tools 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 10d 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 — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.
scvi-tools Deep Learning for Single-Cell Omics
Local, guided workflows for training and applying scvi-tools' probabilistic deep-learning models on a user's own AnnData files.
Honesty contract (read first)
This skill trains real neural networks via the real scvi-tools package. It
must never fabricate a latent space, a training curve, or a DE result.
- Preflight before anything. Every script checks
import scvi(andanndata/scanpy/numpy) before touching data. If scvi-tools is not installed, the script prints the exactpip install scvi-toolscommand and exits 0 — it does not estimate, guess, or describe hypothetical training results. - Raw integer counts required. Every training script verifies the input layer looks like raw counts and refuses to train on normalized/log data with a clear error, rather than letting scvi-tools fail deep inside training or silently produce meaningless output.
- Report the device honestly. Every script prints whether it is training on GPU (CUDA/MPS) or CPU before starting, since CPU training can be 10-50x slower — never claim GPU training happened when it did not.
- Never overwrite the input file. All outputs go to a separate directory/file passed by the caller.
When to Use This Skill
Use when the user:
- Names scVI, scANVI, totalVI, PeakVI, MultiVI, DestVI, veloVI, or scArches specifically
- Wants deep-learning-based batch correction / integration for scRNA-seq with strong or nonlinear batch effects
- Has CITE-seq (RNA + surface protein), multiome (RNA + ATAC), or spatial + single-cell reference data and wants a joint/deconvolved model
- Wants to map a new (query) dataset onto an existing trained reference model (reference mapping / label transfer)
- Wants RNA velocity with model-based uncertainty (not just
scvelo's point estimates)
NOT for (use other skills instead):
- Plain scanpy-based QC, normalization, clustering, marker genes, or
Harmony batch correction with no deep learning involved ->
tooluniverse-single-cell - Running scVI or scANVI as a shared, remote GPU-backed MCP tool for other
callers (not local, guided analysis of your own file) ->
setup-scvi-remote-tool/setup-scanvi-remote-tool - Bulk RNA-seq DESeq2 ->
tooluniverse-rnaseq-deseq2 - FASTQ-level QC/trimming ->
tooluniverse-fastq-qc
What ships with it
10 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.
- references/model_selection.md 3.3 KB
- references/multimodal_workflows.md 4.0 KB
- references/scrna_integration.md 2.8 KB
- references/troubleshooting.md 3.7 KB
- scripts/_scvi_common.py 3.2 KB runs code
- scripts/differential_expression.py 3.3 KB runs code
- scripts/prepare_data.py 3.3 KB runs code
- scripts/train_model.py 6.0 KB runs code
- scripts/transfer_labels.py 3.8 KB runs code
- scripts/validate_adata.py 5.2 KB runs code
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.
- 10d ago First seen · 106 lines · 204 tokens per session scan A 7c48346b4f96
tooluniverse-scvi-tools is a skill published in the GitHub repository mims-harvard/ToolUniverse (1,744 stars, last pushed yesterday), licensed Apache-2.0. It adds 204 tokens to every session and 1,342 once invoked, about $0.0008 per session on Opus 5.5. 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 2026-10-07.
Other skills, from other repositories
experiment-design
Use when the user wants to design experiments, plan ablation studies, structure baselines, or create incremental evaluation strategies. Triggers on phrases like "design ablation", "plan experiment", "what experiments should I run", "baseline comparison", or "experiment matrix".
m4-api
Query M4 clinical datasets (MIMIC-IV, its demo, MIMIC-IV-Note, eICU) from Python with import m4: find tables, run SQL into pandas, search clinical notes, handle M4's errors. Use before writing Python or SQL against M4 data.
mechanism-audit
Audit the mechanistic experiment rigor for a specific claim. Catalogue currently has six slots A–F: A (steering coefficient sweep) is implemented; B–F are reserved for future checks (direction extraction quality, site/layer selection, neffective sufficiency, probe-vs-causal disentanglement, intervention scope). Uses…
dynamic-components
Identify and manipulate language-specific neurons in multilingual Large Language Models (LLMs) to understand and control language-specific behaviors in models like LLaMA-2, BLOOM, OPT, Mistral, and Phi-2.
shap
Use this skill when working with SHAP (SHapley Additive exPlanations) to explain machine learning model predictions, compute feature importance, generate SHAP values for tree ensembles (XGBoost, LightGBM, CatBoost, scikit-learn), deep learning models (TensorFlow, Keras, PyTorch), NLP transformers, or any…
mechanism-skills
Routing entry point for eleven families of mechanistic-interpretability methods that localize which internal object (layer, attention head, neuron, SAE feature, weight, or input feature) drives a model's behavior, how influential it is, and what changes when it is intervened on. Use this skill whenever the question is…