k-dense-ai/scientific-agent-skills
146 skills124k installs5M starsGitHub
Install
npx skills add https://github.com/k-dense-ai/scientific-agent-skillsSkills in this repo
1Scientific Writingscientific-writing is a k-dense-ai scientific-agent-skills module centered on the scientific_report.sty LaTeX package for professional scientific documents. The style applies a Helvetica font family, a blue-green accent color scheme, colored box environments, alternating-row tables, scientific notation commands for p-values and effect sizes, and automatic section headers and footers. Developers reach for scientific-writing when agents must produce consistently formatted scientific reports, technical white papers, or research summaries instead of plain Markdown notes. The skill is a formatting guide for LaTeX deliverables suitable for lab reports, method write-ups, and stakeholder-facing technical papers.1.1kinstalls2Scientific Visualizationscientific-visualization is a Claude Code skill from k-dense-ai/scientific-agent-skills that ships curated color palettes optimized for scientific publications and accessibility. Its color_palettes module exposes the Okabe-Ito palette (2008) with eight named colors—orange, sky_blue, bluish_green, yellow, blue, vermillion, reddish_purple, and black—and helpers such as apply_palette('okabe_ito') for matplotlib.pyplot plots. Developers reach for scientific-visualization when agent-generated figures must meet journal accessibility standards instead of default matplotlib colors that fail colorblind readers. Usage is import-driven: from color_palettes import OKABE_ITO, apply_palette, then plt.plot after apply_palette. The skill fits research-oriented AI agents, thesis pipelines, and reproducible analysis notebooks where figure quality and WCAG-minded contrast matter alongside statistical correctness.1.1kinstalls3Scientific Critical Thinkingscientific-critical-thinking is a k-dense-ai/scientific-agent-skills guide for reducing systematic errors during AI-assisted scientific research. It documents cognitive biases—including confirmation bias, hindsight bias, and publication bias—with concrete manifestations and mitigations such as preregistered hypotheses, blinded analysis, and actively seeking disconfirming evidence. Developers reach for scientific-critical-thinking when agents draft literature reviews, design studies, or evaluate claims where selective citation, ambiguous result interpretation, or post-hoc rationalization could slip in. The skill trains agents to challenge preexisting beliefs and consider alternative hypotheses before research conclusions harden into project direction.1.1kinstalls4Scientific Brainstormingscientific-brainstorming is a scientific-agent-skills reference that supplements standard prompting with structured brainstorming methodologies for technical and scientific ideation. The documented SCAMPER framework provides seven transformation lenses—Substitute, Combine, Adapt, Modify, Put to another use, Eliminate, and Reverse—for improving methods, adapting known techniques, and exploring alternative organisms, systems, or datasets. Developers reach for scientific-brainstorming when a research direction stalls, a PI requests a named methodology, or an agent needs disciplined ideation beyond open-ended chat before designing experiments or grant proposals.1.1kinstalls5Literature Reviewliterature-review is a scientific agent skill for producing structured academic literature reviews before new projects or features. The output template supports narrative, systematic, scoping, meta-analysis, and umbrella review types with PRISMA compliance tracking and optional PROSPERO protocol registration. Sections cover abstract, background, objectives, methods, results with study counts, conclusions, and keywords. Developers and researchers reach for literature-review when they need a standardized, reproducible literature synthesis with explicit selection criteria and quality assessment rather than informal paper summaries.1.1kinstalls6Paper Lookuppaper-lookup is a Claude Code skill from k-dense-ai/scientific-agent-skills for direct arXiv API access from a coding agent. It calls GET https://export.arxiv.org/api/query with search_query, start, and max_results parameters, returning Atom XML responses covering physics, mathematics, computer science, quantitative biology, finance, statistics, electrical engineering, and economics preprints. No authentication is required. Developers reach for paper-lookup when they need arXiv abstracts, paper IDs, or field-prefixed boolean searches without leaving the agent session.1kinstalls7Scientific Schematicsscientific-schematics is a k-dense-ai/scientific-agent-skills guide for creating high-quality scientific diagrams that meet journal requirements and remain accessible to all readers. It specifies vector format rules—PDF for LaTeX line drawings and flowcharts, EPS for legacy systems—and accessibility practices alongside publication standards for fonts, scaling, and embedding. Developers reach for scientific-schematics when preparing block diagrams, circuit drawings, or flowcharts destined for peer-reviewed submission rather than web dashboards. The skill emphasizes scalable vector output, small file sizes, and standards compliance so agents produce camera-ready figures. It bridges research documentation build needs with design rigor typically missing from generic diagram generators.1kinstalls8Scientific Slidesscientific-slides is a scientific agent skill that produces Beamer LaTeX slide decks tuned for academic and research presentations. The template uses a 16:9 aspect ratio, Madrid theme with beaver color theme, amsmath and booktabs packages, graphicx with a figures path, and biblatex with biber for authoryear citations from a references.bib file. Developers and researchers reach for scientific-slides when they need conference-ready slides with proper math typesetting, citation formatting, and consistent visual structure instead of rebuilding Beamer scaffolding by hand. The skill outputs compilable .tex sources with frame numbers, stripped navigation symbols, and customizable primary and secondary colors. scientific-slides fits paper talks, lab meetings, and thesis defenses where LaTeX quality matters.1kinstalls9Matplotlibmatplotlib is skill version 1.1 from k-dense-ai/scientific-agent-skills, authored by K-Dense Inc., guiding AI agents through Python's Matplotlib 3.10.x library for static, animated, and interactive visualizations. The skill documents both pyplot and recommended object-oriented workflows, covering subplots, GridSpec, mosaic layouts, plot types from line and scatter to heatmaps and violins, styling, colormaps, and dpi-controlled exports. Setup uses uv add matplotlib with optional ipympl for Jupyter widgets, requiring Python 3.10+ and NumPy 1.23+. Agents reach for matplotlib when fine-grained control beats seaborn shortcuts or plotly interactivity, especially for multi-panel publication figures. The skill points to references/plot_types.md for extended examples and notes scientific-visualization for journal-styled figures. Catalog data shows 792 installs for this entry among scientific agent skills.1kinstalls10Citation Managementcitation-management is a scientific-agent-skills workflow for producing and maintaining BibTeX bibliographies in research-oriented projects. It ships template entries for common source types including journal articles, with fields such as author, title, journal, volume, number, pages, and year demonstrated on references like AlphaFold in Nature. The skill helps agents and developers keep .bib files consistent, properly escaped, and ready for LaTeX or Pandoc pipelines instead of hand-typing references that break builds. Use it when technical docs, ML papers, or agent knowledge bases need durable citation files and formatted reference lists.1kinstalls11Statistical Analysisstatistical-analysis is a k-dense-ai scientific-agent-skills module for assumption checking and diagnostic procedures across common statistical tests. It instructs agents to verify independence, normality, homoscedasticity, and related conditions using visual and formal methods before interpreting results. The skill requires documenting violations and remedial actions in analysis reports. Developers and data scientists reach for statistical-analysis when regression, ANOVA, or inference code might run without diagnostics, risking false conclusions in experiments, A/B tests, or research pipelines.1kinstalls12Peer Reviewpeer-review is a K-Dense scientific agent skill that produces structured peer reviews using checklist-based evaluation of methodology, statistical validity, and reporting standards such as CONSORT and STROBE. The skill is MIT-licensed and allows Read, Write, Edit, and Bash tools so agents can draft constructive formal feedback on manuscripts and grant proposals. Use peer-review when writing an actual review document with specific criteria rather than loose claim checking. For evidence-quality critique, the repo points to scientific-critical-thinking; for quantitative scoring, use scholar-evaluation.1kinstalls13Pdfpdf is a k-dense-ai scientific-agent-skills module that reliably reads, analyzes, and fills PDF forms through an ordered agent workflow. The skill requires checking fillable fields first with `python scripts/check_fillable_fields <file.pdf>`, then branching to fillable or non-fillable field paths. For fillable PDFs, `python scripts/extract_form_field_info.py <input.pdf> <field_info.json>` emits a JSON list of field_id, page, and metadata for programmatic filling. Developers reach for pdf when automating government, scientific, or compliance forms where manual Acrobat edits or fragile OCR would break agent pipelines.1kinstalls14Scikit Learnscikit-learn is a scientific-agent-skills reference that walks coding agents through scikit-learn model selection workflows including train-test splits, stratified sampling, three-way train/val/test splits, cross-validation, grid search, and metric-driven model comparison. The skill encodes Python patterns for train_test_split with random_state control, stratify for imbalanced classes, and validation pipelines before agents deploy classifiers or regressors into production data flows. Developers reach for scikit-learn when an agent must compare algorithms, tune hyperparameters, or document evaluation methodology instead of shipping an untested estimator. The guide targets sklearn.model_selection APIs and pairs naturally with pandas feature matrices and sklearn estimators in agent-driven notebooks or CI benchmarks.1kinstalls15Markdown Mermaid Writingmarkdown-mermaid-writing is a skill from k-dense-ai/scientific-agent-skills that produces scientific, technical, and research documents where Mermaid diagrams are embedded directly in markdown as the source format. The bundled example report on CRISPR-Cas9 gene editing demonstrates standards: overview sections, quantified findings such as 78% mean editing efficiency in HEK293T cells, correlation notes, and diagrams rendered from Mermaid blocks. Developers reach for markdown-mermaid-writing when turning analysis results, lab workflows, or system designs into shareable reports that need both narrative prose and flowcharts, sequence diagrams, or process charts without separate drawing tools. Output targets publication-ready structure rather than casual notes.985installs16Exploratory Data AnalysisExploratory Data Analysis is a scientific-agent-skills workflow that produces a structured EDA report template for a given file before modeling or visualization work begins. Reports include an executive summary, basic file metadata such as filename, full path, byte size, last modified date, extension, and format category, plus format-specific descriptions and typical data content sections. The skill also covers data quality assessment and recommendations for downstream analysis paths. Developers reach for Exploratory Data Analysis when onboarding unfamiliar CSV, JSON, or scientific datasets and need a consistent pre-modeling audit. The generated report gives agents and engineers a shared baseline on schema, anomalies, and next analytical steps.981installs17Docxdocx is a k-dense-ai scientific-agent-skills package, licensed under Anthropic terms, that equips agents to manipulate Microsoft Word .docx files with accurate structured reading, editing, and generation. Developers use docx when reports, protocols, or scientific write-ups must live in Word format rather than plain Markdown, especially in regulated or collaborator-facing workflows. The skill fits agent pipelines that ingest existing .docx templates, patch sections programmatically, and emit updated Word documents without manual copy-paste between editors.978installs18Pptxpptx is a k-dense-ai scientific-agent-skills agent skill for template-based PowerPoint generation that avoids monotonous title-plus-bullet decks. The workflow starts by analyzing an existing template.pptx with python scripts/thumbnail.py to produce thumbnails.jpg and python -m markitdown to extract placeholder text, then maps each content section to a distinct template slide layout. The skill actively recommends multi-column, image-plus-text, full-bleed, and quote callout layouts instead of defaulting to basic bullets. Developers reach for pptx when an agent must produce presentation files from structured research or report content while preserving branded template styling.977installs19Seabornseaborn is a K-Dense scientific-agent-skills module (skill metadata version 1.1, BSD-3-Clause) for publication-quality statistical graphics built on matplotlib with pandas integration. Official examples target seaborn 0.13.2 on Python 3.8+ with mandatory NumPy, pandas, and matplotlib; optional scipy, statsmodels, and fastcluster install via seaborn[stats]==0.13.2. The skill documents both the traditional function interface—scatterplot, displot, pairplot, heatmaps—and the experimental seaborn.objects composable API similar to ggplot2 for declarative plot building. Agents use it for distribution exploration, categorical comparisons, box and violin plots, and multi-panel faceted figures saved at dpi=300 with bbox_inches=tight. Reach for seaborn over plotly or interactive dashboards when static publication PNGs from DataFrames are the deliverable. sns.load_dataset fetches public examples when cached; private regulated or offline work should load local CSV files through pandas read_csv before plotting with agents in Bash-enabled Python 3.8+ sessions.974installs20Database Lookupdatabase-lookup is a k-dense-ai/scientific-agent-skills reference for querying Addgene and AlphaFold without leaving an agent workflow. For Addgene, it documents the https://www.addgene.org/api/ base URL, Token authorization via ADDGENE_API_KEY in .env, and endpoints for plasmids by ID, plasmid search, depositors, and articles—with example GET calls such as /plasmids/12260/ for pSpCas9. Developers reach for database-lookup when agents need authoritative plasmid metadata or structure predictions instead of hallucinated accession details. The skill emphasizes API key setup and concrete HTTP patterns agents can execute programmatically.971installs21Markitdownmarkitdown is a K-Dense scientific-agent-skills module (metadata version 1.1, MIT license) wrapping Microsoft's MarkItDown Python converter for agent-friendly document ingestion. It supports 15+ input formats including PDF, DOCX, PPTX, XLSX, HTML, CSV, JSON, XML, ZIP, EPUB, images with EXIF/OCR, audio with transcription, and YouTube URLs. The MarkItDown class exposes convert() with auto-detection plus convert_local() for filesystem paths, producing token-efficient markdown suited to LLM prompts and RAG chunking. Optional OPENROUTER_API_KEY enables LLM-powered image descriptions. Basic usage is a few lines: instantiate MarkItDown(), call convert(path), and write result.text_content to output.md. Reach for markitdown when you need reliable office-to-markdown conversion inside scientific or engineering agent workflows instead of manual copy-paste or brittle pandoc one-offs.970installs22Statsmodelsstatsmodels is a K-Dense scientific-agent-skills module (metadata version 1.1, BSD-3-Clause) for Python statistical modeling with statsmodels 0.14.6. It covers OLS and WLS regression, generalized linear models, binary and multinomial logit, Poisson and count models including hurdle specifications, ordinal outcomes, and ARIMA/SARIMAX time series with AIC/BIC model comparison. Examples emphasize sm.add_constant for intercepts, results.summary() tables, Breusch-Pagan heteroskedasticity tests, residual plots, and publication-ready inference. Install via uv pip install statsmodels==0.14.6 on Python 3.9+. Reach for statsmodels when you need explicit model classes with residual diagnostics and coefficient inference—use the separate statistical-analysis skill instead for guided test selection with APA reporting.967installs23Hypothesis GenerationHypothesis-generation is a k-dense-ai scientific agent skill that produces structured hypothesis reports using a dedicated LaTeX template and `hypothesis_generation` style package. Reports distinguish multiple hypotheses with a defined color scheme, supporting predictions, evidence comparisons, and stated limitations. Compilation uses XeLaTeX or LuaLaTeX with a standard `xelatex → bibtex → xelatex × 2` workflow and natbib citations. Researchers and computational scientists reach for hypothesis-generation when they need agent-assisted brainstorming turned into publication-ready, visually differentiated hypothesis documents rather than unstructured bullet lists. The bundled quick-reference guide covers documentclass setup, compilation commands, and per-hypothesis color assignments.959installs24Scholar Evaluationscholar-evaluation implements the ScholarEval evaluation framework from k-dense-ai/scientific-agent-skills for systematic scholarly quality review. The skill supplies dimension-specific rubrics—starting with Problem Formulation and Research Questions—each scored on a 5-point scale where Excellent (5) demands specific measurable questions, significant literature gaps, appropriate scope, clear novelty, and compelling significance justification. Developers reach for scholar-evaluation when vetting research proposals, paper drafts, or LLM-generated academic content before peer review or publication. Agents apply the documented quality indicators and rubric tiers to produce structured, comparable evaluation scores across scholarly dimensions.959installs25Bgpt Paper Searchbgpt-paper-search is a scientific-agent skill (version 1.1) that searches papers through the BGPT MCP server and returns 25+ structured fields per study, including methods, results, sample sizes, quality scores, and conclusions extracted from full text. It requires the BGPT MCP server configured in the agent host via npx mcp-remote or npx bgpt-mcp, internet access to bgpt.pro, and an optional BGPT API key for paid usage. Developers reach for bgpt-paper-search during literature reviews, evidence synthesis, or when they need experimental details missing from PubMed abstracts. The skill fits agent workflows that must cite quantitative findings with reproducible field-level data rather than summary paraphrases.956installs26Pptx Posterspptx-posters is a K-Dense scientific-agent skill that builds research posters with modern HTML/CSS layouts, responsive sections, and straightforward visual integration. The skill exports finished posters directly to PDF or converts them into editable PowerPoint (PPTX) format when a developer explicitly requests that output. The manifest restricts tooling to Read, Write, Edit, and Bash under an MIT license, and the readme directs agents to prefer latex-posters for standard research posters while reserving pptx-posters for PPTX-specific requests. Developers reach for pptx-posters when conference or lab poster requirements mandate PowerPoint handoff, rapid web-based iteration, or PDF export without a LaTeX toolchain.944installs27Latex Posterslatex-posters is a skill from k-dense-ai/scientific-agent-skills that instantly generates publication-ready LaTeX research posters and academic visuals using the baposter document class. The bundled template uses a0paper portrait layout with fontscale=0.285 and packages including graphicx, amsmath, booktabs, multicol, qrcode, hyperref, and enumitem for professional multi-column poster design. Developers reach for latex-posters when preparing conference posters, lab presentation boards, or academic symposium materials that need structured baposter layouts with automatic column positioning. The template comments note baposter excels at multi-column layouts with automatic positioning for research symposium deliverables.937installs28Xlsxxlsx is an Anthropic-licensed spreadsheet skill from k-dense-ai/scientific-agent-skills that enables agents to read, write, and process Excel .xlsx workbooks within Claude or Cursor sessions. It targets scientific and engineering workflows where tabular data must be parsed, transformed, or exported without leaving the agent environment. Developers reach for xlsx when automating workbook generation, extracting structured data from Excel files, or building data pipelines that interchange .xlsx artifacts with code and analysis scripts.933installs29Research Grantsresearch-grants is a scientific-agent-skills template for detailed budget justifications that explain each line item's necessity, calculation, and tie to proposed research. Sections cover personnel salaries (including PI effort), with principles to justify every item, show cost-effectiveness, and follow agency-specific formats. Developers in research engineering or lab software roles reach for research-grants when translating a technical research plan into reviewer-ready budget narrative instead of ad hoc spreadsheet notes.929installs30Generate Imagegenerate-image is a scientific-agent skill from k-dense-ai/scientific-agent-skills shipping a Python CLI for OpenRouter-backed image generation and editing. Supported models include google/gemini-3.1-flash-image-preview, black-forest-labs/flux.2-pro, and black-forest-labs/flux.2-flex, with editing via an input image plus prompt. The script checks for OPENROUTER_API_KEY in environment or .env files and outputs generated or edited image files. Developers reach for generate-image when agents or pipelines need reproducible image creation without wiring each model SDK manually.922installs31PolarsPolars is a performance guide skill from k-dense-ai/scientific-agent-skills for writing efficient Polars Python code. It prioritizes lazy evaluation with scan_csv and collect so queries benefit from predicate pushdown, projection pushdown, query optimization, and parallel execution planning instead of eager full-file loads. The skill teaches pushing filter and select operations early in pipelines, using the expression API correctly, and avoiding common eager-mode pitfalls that stall large datasets. Developers invoke Polars when agents generate dataframe code that works on samples but will choke on production CSV or Parquet volumes. It complements scientific and analytics workflows where Polars replaces pandas for speed-critical transforms. The readme contrasts bad eager patterns with optimized lazy pipelines using concrete Python examples developers can apply immediately.922installs32Pytorch Lightningpytorch-lightning is a skill from k-dense-ai/scientific-agent-skills that enforces clean separation between research logic and engineering boilerplate when training deep learning models with PyTorch Lightning. It documents good patterns—training_step and compute_loss inside L.LightningModule while L.Trainer handles max_epochs, GPU accelerator, devices, and ddp strategy—and flags anti-patterns like manual device management inside training_step. Developers reach for pytorch-lightning when refactoring messy training scripts, setting up multi-GPU DDP runs, or onboarding agents to write maintainable Lightning code where research experiments stay isolated from distributed training configuration. The skill targets scientific ML workflows where reproducibility and code organization matter as much as model architecture.915installs33Aeonaeon is an agent skill from k-dense-ai/scientific-agent-skills that documents Aeon anomaly detection for time series at both single-series and collection levels. Collection detectors include ClassificationAdapter for training on normal data and flagging outliers, plus OutlierDetectionAdapter wrapping sklearn detectors such as IsolationForest, LOF, and OneClassSVM. Series detectors identify anomalous points or subsequences within individual series. Developers reach for it when building monitoring agents, scientific data pipelines, or analytical tools that must surface unusual patterns without hand-rolling statistical tests. The skill guides method selection—for example classification-based approaches with labeled normal data versus sklearn outlier wrappers on feature collections.913installs34Networkxnetworkx is a skill from k-dense-ai/scientific-agent-skills that embeds Python NetworkX graph algorithm recipes directly into Claude-powered agent workflows. The reference covers single-source shortest paths via Dijkstra and Bellman-Ford, all-pairs shortest paths, connectivity analysis, and centrality measures with copy-paste code snippets such as nx.shortest_path, nx.bellman_ford_path, and nx.all_pairs_shortest_path. Developers reach for networkx when building agents or scripts that must analyze weighted graphs, find optimal routes, or measure node importance without leaving the agent context. The skill targets scientific and data engineering workloads where graph topology drives recommendations, dependency resolution, or network diagnostics.911installs35Get Available Resourcesget-available-resources is a skill from k-dense-ai/scientific-agent-skills built around a Python system resource detection script. The script inspects CPU topology, GPU availability, memory, and disk space, then writes JSON Claude Code can consume to pick computational approaches—whether to use Dask, Zarr, Joblib, or GPU-backed stacks. Platform support spans macOS, Linux, and Windows. GPU detection covers NVIDIA CUDA, AMD ROCm, and Apple Silicon Metal backends. Functions such as get_cpu_info use psutil and subprocess probes to assemble machine facts agents would otherwise guess incorrectly. Developers reach for it before launching notebooks, batch jobs, or agent-driven data pipelines where wrong parallelism wastes memory or fails on missing GPUs. Run it early in scientific workflows so downstream library choices match real hardware.909installs36TransformersTransformers is a text-generation skill from k-dense-ai/scientific-agent-skills built around the Hugging Face Transformers library. It documents generating text with model.generate(), controlling output through generation strategies and parameters, and choosing between the Pipeline API for quick prototyping versus direct AutoModelForCausalLM and AutoTokenizer usage for custom preprocessing and decoding control. Examples use gpt2 with AutoModelForCausalLM.from_pretrained, tokenizer input handling, and max_new_tokens generation. Developers reach for Transformers when building scientific agents or Python automation that needs fine-grained control over LM decoding rather than opaque API calls. The skill bridges Pipeline convenience and low-level generate() customization.908installs37Infographicsinfographics is a reference skill from k-dense-ai/scientific-agent-skills providing comprehensive color palette options for accessible, professional infographics. The guide centers on colorblind-safe palettes including Wong's Palette with 7 colors—Black, Orange, Sky Blue, Bluish Green, and additional distinguishable hues developed by Bang Wong for scientific visualization—plus IBM and other accessible hex palettes with documented RGB values and usage notes. Developers and agents reach for infographics when generating charts, diagrams, or data visualizations that must remain distinguishable for viewers with color vision deficiency. The skill supplies hex codes, RGB tuples, and usage guidance rather than rendering pipelines, making it a design reference layer for scientific content generation.907installs38Paperzillapaperzilla is a skill from k-dense-ai/scientific-agent-skills at version 1.0 by Paperzilla Inc (MIT licensed) for chatting with agents about Paperzilla projects, recommendations, and canonical research papers. Developers can request latest recommendations from a project, open a recommendation with rationale, fetch canonical papers as markdown summaries, submit recommendation feedback, or export feeds including Atom feed URLs. Example prompts include asking for canonical paper markdown, explaining why a recommendation matters, and pulling structured feed exports for literature review during implementation. Reach for paperzilla when scientific or ML features need grounded citations instead of hallucinated references. The skill bridges Paperzilla's recommendation graph and canonical paper store into Claude Code sessions so literature context stays tied to identifiable project artifacts and exportable feeds.907installs39Pyzoteropyzotero is a scientific-agent skill that teaches coding agents to interact with personal or group Zotero libraries through the official API. It documents credential setup from zotero.org/settings/keys, environment-variable handling for ZOTERO_* keys, and safe patterns that avoid hardcoding secrets in source control. Developers use it when building automated literature review, bibliography sync, or research pipelines that search, fetch, and update Zotero items. The skill emphasizes security-first API key management and correct user ID discovery for API calls, making agent-generated pyzotero code production-safe for academic and R&D teams.907installs40Optimize For Gpuoptimize-for-gpu is a scientific computing skill that guides agents to transform CPU-bound Python—loops, large arrays, ML pipelines, graph analytics, and image processing—into GPU code via 12 NVIDIA libraries including CuPy, Numba CUDA, Warp, cuDF, cuML, cuGraph, KvikIO, cuCIM, cuxfilter, cuVS, cuSpatial, and RAFT. The skill covers physics simulation, particle systems, geospatial analysis, medical imaging, vector search, and sparse eigensolvers when users mention GPU, CUDA, or NVIDIA acceleration. Developers reach for optimize-for-gpu when pandas, scikit-learn, NetworkX, GeoPandas, or Faiss code is too slow on CPU and needs production-grade GPU equivalents without rewriting algorithms from scratch.906installs41Market Research Reportsmarket-research-reports is a scientific-agent-skills module that applies the `market_research.sty` LaTeX style package to produce visually structured market research PDFs. The style guide defines a primary blue palette (`primaryblue` #003366, `secondaryblue` #336699, `lightblue` #ADD8E6, `accentblue` #0078D7) plus accent greens for market data boxes and positive indicators. Developers reach for this skill when agents must output investor-ready or stakeholder-ready market analyses with repeatable section styling, key insight boxes, and opportunity highlights instead of ad-hoc Markdown exports. It focuses on document formatting conventions rather than raw data collection.904installs42Shapshap is a k-dense-ai scientific agent skill that documents the full SHAP explainer reference for Python ML workflows. The general shap.Explainer class auto-selects the best algorithm for a given model type, while specialized explainers optimize for tree models, deep networks, linear models, and arbitrary functions. Developers reach for shap when they need interpretable Shapley-value attributions instead of black-box predictions—for compliance reviews, feature audits, or debugging model behavior. The skill covers constructor parameters, method usage, and selection guidance across SHAP explainer classes so agents pick the correct explainer rather than defaulting to a slow or inaccurate option.900installs43Sympysympy is an agent skill from the k-dense-ai scientific-agent-skills collection that teaches coding agents to produce correct SymPy Python for symbolic mathematics. Coverage spans advanced topics including 2D geometry with Point, Line, Circle, Triangle, and Polygon objects, number theory, combinatorics, logic and sets, statistics, polynomials, and special functions. Developers reach for sympy when building scientific calculators, engineering tooling, or agent workflows that need exact symbolic solutions rather than floating-point approximations. The skill provides code patterns for distance calculations, line equations, segment lengths, and equation systems so generated Python executes without common SymPy API mistakes.900installs44Daskdask is a scientific-agent skill explaining Dask Array, which implements NumPy's ndarray interface with blocked algorithms. Data splits into chunks where each chunk is a regular NumPy array, operations run in parallel across cores, and results combine automatically for out-of-core workloads. The reference covers supported mathematical operations including arithmetic, exponentials, logarithms, and trigonometric functions applied per chunk. Developers reach for it when pandas or raw NumPy runs out of memory on large scientific arrays and they need familiar ndarray APIs with transparent parallelism and disk-scale datasets.899installs45Matlabmatlab is a scientific-agent skill that gives coding agents a structured reference for MATLAB data import and export. It covers six areas: text and CSV files, spreadsheets, MAT files, images, tables and data types, and low-level file I/O. Examples show readtable for mixed-type tables, readmatrix for numeric data, readcell for cell arrays, readlines for line-based text, and fileread for whole-file strings. Developers reach for it when agents must generate correct high-level I/O instead of brittle low-level fopen patterns. The skill targets reproducible scientific pipelines where file format choice affects downstream matrix and table operations.896installs46Open Notebookopen-notebook is a scientific-agent skill that integrates agents with the Open Notebook REST API at http://localhost:5055/api, with Swagger UI at /docs and ReDoc at /redoc. It covers notebook CRUD, archival filters, sorting, and optional password auth via OPEN_NOTEBOOK_PASSWORD on protected routes. Agents use it to read prior findings, append structured notes, and organize research across sessions instead of losing context in chat history. Developers reach for open-notebook on literature reviews, experiment logs, or any project needing durable, queryable notebooks the agent can manage programmatically.893installs47Parallel Webparallel-web is a Claude agent skill from k-dense-ai/scientific-agent-skills (version 1.1, 693 catalog installs) that unifies parallel-cli web capabilities inside coding sessions. The SKILL.md routes requests to web search, URL extraction, CSV or inline JSON data enrichment, and deep research, defaulting to academic sources like arXiv, bioRxiv, and peer-reviewed journals for technical queries. Developers invoke Bash commands such as parallel-cli enrich run with --source-type csv, parallel-cli research status, and parallel-cli research poll, using --no-wait for non-blocking async jobs and --previous-interaction-id to chain follow-up questions. Setup installs parallel-cli via curl or uv and authenticates with PARALLEL_API_KEY from a .env file or parallel-cli login. Reach for parallel-web when you need batch company field enrichment, literature lookups, PDF extraction, or exhaustive multi-source reports without leaving the agent workflow.893installs48Scikit Survivalscikit-survival is a scientific-agent skill for competing-risks analysis where subjects face several mutually exclusive events—death from cancer versus cardiovascular disease, relapse versus death, or equipment failure modes. It explains why standard Kaplan-Meier breaks down and how cumulative incidence functions estimate event-type probabilities when one outcome blocks others. The skill walks through scikit-survival patterns for medical, reliability, and churn studies. Developers reach for scikit-survival when survival endpoints have competing events and they need correct incidence curves instead of naive single-event survival estimates.892installs49Pymcpymc is a workflow template from k-dense-ai/scientific-agent-skills for Bayesian hierarchical (multilevel) models using PyMC, ArviZ, NumPy, pandas, and matplotlib. The template walks through numbered sections for data preparation on grouped structure such as students within schools or patients within hospitals, model definition with explicit TODO customization markers, MCMC sampling, ArviZ diagnostic review, and matplotlib plotting hooks. Developers reach for pymc when grouped data needs partial pooling instead of flat frequentist models, and they want a complete Python scaffold rather than assembling PyMC documentation piecemeal. Each section preserves a standard hierarchical modeling sequence from CSV load through inference while leaving hooks for project-specific priors and predictors. pymc suits scientific and backend developers building reproducible Bayesian pipelines where group_id columns define nested experimental units. The skill delivers runnable starter code agents can extend rather than abstract theory alone.885installs50Torch Geometrictorch-geometric is a K-Dense scientific-agent-skills reference (skill version 1.1) for PyTorch Geometric 2.7.x on Python 3.10+ and PyTorch 2.6+. It documents 60+ GNN conv layers, Data and HeteroData objects, Planetoid and TUDataset loaders, and four InMemoryDataset overrides for reusable graph datasets from CSV, pandas, numpy, NetworkX, or scipy sparse matrices. The skill covers node, graph, and link prediction patterns, NeighborLoader scaling with num_neighbors aligned to GNN depth, heterogeneous to_hetero models, and custom MessagePassing layers with explicit edge_index COO format rules. Optional accelerated wheels from data.pyg.org are documented but not required for basic usage since PyG 2.3. Developers reach for torch-geometric when agents must build or debug graph ML pipelines—loading proprietary edge tables, choosing GCN versus GAT versus GraphSAGE, or avoiding pitfalls like wrong edge_index orientation or missing activations after conv layers.882installs51Umap Learnumap-learn is a K-Dense scientific-agent-skills reference (skill version 1.1) for umap-learn 0.5.12 on Python 3.9+ with scikit-learn 1.6 or newer. It documents four primary parameters—n_neighbors, min_dist, n_components, and metric—with tuning recipes for visualization, clustering prep, and document cosine embeddings. Workflows cover StandardScaler preprocessing, supervised and semi-supervised UMAP with y labels marked -1 for unlabeled rows, HDBSCAN clustering at n_components 5–10 with min_dist 0.0, transform on held-out test data, and Parametric UMAP via TensorFlow extras installable as umap-learn[parametric-umap]==0.5.12. Advanced sections cover AlignedUMAP for temporal batches, inverse_transform reconstruction limits, and DensMAP variants. The skill warns against shadowing imports with local umap.py files and recommends random_state=42 for reproducibility. Developers reach for umap-learn when agents must choose UMAP settings over t-SNE or PCA, wire embeddings into sklearn pipelines, or debug disconnected components and collapsed clusters.882installs52Hypogenichypogenic is a skill from k-dense-ai/scientific-agent-skills that configures HypoGeniC-style hypothesis generation and testing over train, validation, and test JSON datasets. It documents model settings for GPT-4, Claude-3, or GPT-3.5-turbo, optional Redis caching on localhost port 6832 to cut API costs, and generation parameters like temperature 0.7 and max_tokens 2048. Developers reach for hypogenic when they need an agent to propose and evaluate scientific hypotheses from labeled text features rather than hand-writing experiment loops. The bundled configuration template specifies dataset schemas with text_features_n lists and label fields for repeatable LLM-driven research runs.878installs53Venue Templatesvenue-templates is a Documentation agent skill in k-dense-ai/scientific-agent-skills that generates Cell Press-specific manuscript elements including Summary abstracts capped at 150 words, Highlights bullet lists, and eTOC blurbs. The skill ships complete worked examples such as a senescence and aging paper showing FOXO4-p53 interactions, giving agents concrete length and tone references before drafting new copy. Developers reach for venue-templates when a Cell Press submission needs venue-accurate Summary, Highlights, and eTOC text instead of generic abstract prose. The skill encodes Cell Press formatting constraints so agents produce submission-ready sections without manually cross-checking author guidelines. Use it alongside broader scientific-writing skills when polishing final journal packages for Cell Press titles.878installs54Biopythonbiopython is a skill from k-dense-ai/scientific-agent-skills that teaches agents advanced Biopython patterns for biological computing. It covers Bio.motifs creation from sequence instances, consensus and degenerate IUPAC sequences, position weight matrices, and genomics-oriented sequence analysis APIs. Developers reach for biopython when an agent must draft motif discovery, PWM calculations, or sequence manipulation code instead of generic Python stubs. The skill includes concrete examples—creating motifs from seven Seq instances like TACAA and TACGC—and accessing counts matrices and pwm objects for downstream genomics workflows.876installs55Consciousness CouncilConsciousness Council is an advanced configuration guide for multi-persona agent deliberations from the scientific-agent-skills repository. Instead of accepting one LLM reply, developers assign domain-specific council members—such as a five-member Startup Decisions panel (Strategist, Pragmatist, Contrarian, Futurist, Empiricist) or a Technical Architecture panel (Architect, Minimalist, Empiricist, Outsider, Pragmatist)—to debate trade-offs like ambition versus execution capacity. The skill documents why each persona mix fits a decision type and which tensions to watch during deliberation. Reach for Consciousness Council when a coding agent must stress-test product, technical, hiring, or creative choices before locking scope.876installs56Scikit Bioscikit-bio is a Claude Code skill from k-dense-ai/scientific-agent-skills that equips Python-based AI agents with production-grade biological sequence analysis, phylogenetic computations, and statistical ecology tools through the scikit-bio library. The skill's API reference covers nine domains: sequence classes, alignment methods, phylogenetic trees, diversity metrics, ordination, statistical tests, distance matrices, and file I/O, with worked examples for DNA, RNA, and Protein sequence objects. Developers reach for scikit-bio when building agents that parse FASTA data, compute phylogenetic trees, run diversity analyses, or perform ordination on ecological datasets inside Python pipelines. The skill includes troubleshooting guidance for common scikit-bio API errors so agents produce reproducible bioinformatics outputs rather than hallucinating method signatures. scikit-bio fits scientific agent builds where biological data processing must run in-process alongside LLM reasoning.875installs57What If Oraclewhat-if-oracle is a k-dense-ai scientific-agent-skills guide for structured what-if analysis using domain-specific scenario templates. It ships configurations for startup and business decisions—market timing, pricing, partnerships, hiring, funding—and technology architecture choices including stack selection, build versus buy, scaling, security models, and migrations. Each template defines variables to test, branch emphasis such as Likely Case, Contrarian, and Second Order effects, and prompt scaffolds with timeframe and constraint placeholders. Developers and tech leads reach for what-if-oracle when comparing irreversible architecture or product bets and need consistent branching instead of ad-hoc brainstorming. The skill encodes tension notes like speed versus thoroughness so agents produce comparable scenario writeups.872installs58Timesfm Forecastingtimesfm-forecasting is a Data Science & ML agent skill from k-dense-ai/scientific-agent-skills that wraps Google's TimesFM foundation model for zero-shot univariate time-series forecasting and two-phase anomaly detection. The bundled example runs Phase 1 linear detrend plus Z-score scoring on 36 months of NOAA temperature anomaly data (2022-01 through 2024-12), then Phase 2 TimesFM quantile prediction intervals on a 12-month synthetic horizon with 3 injected anomalies. The workflow writes a 2-panel matplotlib chart to output/anomaly_detection.png and structured detection records to output/anomaly_detection.json. Developers reach for timesfm-forecasting when they need probabilistic forecasts, calibrated quantile intervals, or anomaly flags on sales, sensor, weather, or vitals series from CSV, DataFrame, or array inputs. The skill includes a mandatory preflight checker to verify RAM, GPU, and disk before loading the model, and works with Cursor, Claude Code, Codex, and other Agent Skills hosts.871installs59Geopandasgeopandas is a skill from k-dense-ai/scientific-agent-skills focused on coordinate reference systems and geospatial data handling in Python. It explains pyproj.CRS objects, detecting missing CRS metadata, using set_crs() when coordinates are correct but untagged, and to_crs() for true reprojection between systems such as EPSG:4326. The readme warns that set_crs() does not transform coordinates—only metadata assignment. Developers reach for this skill when agents must correctly load, transform, and visualize geospatial datasets without CRS mistakes that silently misplace features.869installs60Stable Baselines3stable-baselines3 is a scientific-agent-skills reference from k-dense-ai that maps every major Stable Baselines3 trainer to action-space fit, sample efficiency, and training speed. Its comparison table covers on-policy options like PPO and A2C plus off-policy continuous-control algorithms SAC and TD3, with guidance for discrete versus continuous action spaces and prototyping versus sample-efficient runs. Developers reach for stable-baselines3 when starting a Gymnasium or custom RL environment and unsure whether PPO's stability, SAC's continuous control, or DQN-style discrete policies fit best. The skill prevents costly wrong-algorithm trials by encoding algorithm type, speed, and use-case rows agents can cite before writing training loops.869installs61Simpysimpy is a version 1.0 agent skill from K-Dense Inc. covering SimPy discrete-event simulation in Python using generator-based processes, shared resources, and event-driven scheduling. The skill documents seven resource types—Resource, PriorityResource, PreemptiveResource, Container, Store, FilterStore, and PriorityStore—plus three common patterns for customer-server queues, producer-consumer flows, and parallel task execution. Bundled scripts include basic_simulation_template.py and resource_monitor.py with CSV export, supported by five reference guides on resources, events, process interaction, monitoring, and real-time simulation. Developers reach for simpy when modeling queue wait times, server utilization, capacity planning, or network traffic where entities compete for shared resources over irregular time intervals.866installs62Clinical ReportsClinical Reports is an agent skill from k-dense-ai/scientific-agent-skills that templates full clinical case report manuscripts for medical and research developers. Output follows standard case report structure: descriptive title including "Case Report" or "Case Study," author affiliations with ORCID IDs, 2–5 MeSH keyword terms, structured abstract sections (introduction, patient concerns, diagnosis), and narrative clinical sections. Developers reach for Clinical Reports when drafting journal-submission-ready documentation for unusual presentations or rare diagnoses, replacing blank-document authoring with agent-guided section filling that preserves academic formatting expectations for clinical publication workflows.861installs63Pymoopymoo is a Claude Code skill that acts as a comprehensive reference for pymoo optimization algorithms in Python scientific and backend workflows. It documents single-objective methods—Genetic Algorithm, Differential Evolution, Particle Swarm Optimization, and CMA-ES—with parameter guidance such as default population size 100, sampling strategies, crossover operators like SBX, and polynomial mutation settings. Developers reach for pymoo when hyperparameters, engineering constants, or multi-objective trade-offs need evolutionary search instead of manual grid tuning. The skill helps pick algorithm families for continuous, discrete, or mixed-variable problem shapes.861installs64Bioservicesbioservices is a scientific agent skill from k-dense-ai that teaches coding agents how to map biological identifiers across heterogeneous databases. Biological pipelines routinely mix UniProt protein IDs, UniChem compound keys, and KEGG pathway identifiers—BioServices documents UniProt Mapping Service workflows, UniChem compound mapping, KEGG identifier conversions, and common cross-reference patterns with troubleshooting guidance. Developers reach for bioservices when agent-generated bioinformatics code must convert IDs between databases without silent mismatches, whether building annotation pipelines, metabolomics workflows, or protein-compound linkage scripts.860installs65Anndataanndata in k-dense-ai/scientific-agent-skills documents best practices for Python AnnData objects used in single-cell genomics pipelines inside AI agents. The skill shows how to detect sparsity, convert dense arrays to scipy csr_matrix or csc_matrix when more than 50% of values are zero, and notes AnnData 0.12+ requirements for sparse dtypes. It also covers converting string metadata columns to categoricals to cut memory use, citing 10–100× reductions for sparse genomics data. Developers reach for anndata when agent-assisted notebooks or services must load large expression matrices without exhausting RAM or when naive dense conversions make downstream scanpy-style workflows unusable.858installs66Geomastergeomaster is a comprehensive geospatial science skill from k-dense-ai/scientific-agent-skills covering remote sensing, GIS, spatial statistics, and ML for Earth observation. The bundle includes 70+ documentation sections, 500+ code examples across 7 programming languages, and references to 300+ geospatial libraries and tools. Core references document GDAL, Rasterio, Fiona, Shapely, PyProj, and GeoPandas, plus satellite optical, SAR, and hyperspectral analysis workflows. Developers reach for geomaster when building spatial ETL pipelines, land-cover classification, or satellite imagery processing inside AI coding agents. It suits data engineers and geospatial scientists who need library-specific examples rather than generic GIS theory. The skill spans installation, quick start, common operations, and multi-step spatial workflows.858installs67Ggetgget is a skill from k-dense-ai/scientific-agent-skills that brings the gget Python toolkit into AI agent sessions for genomic, protein, and reference sequence lookups. Its database directory documents sources such as Ensembl—used by gget ref, gget search, gget info, and gget seq modules—for annotated vertebrate and invertebrate genomes with ongoing updates. Because upstream databases change structure, gget modules are tested automatically on a biweekly basis and updated to match new schemas; the docs recommend `pip install --upgrade gget`. Developers reach for gget when bioinformatics pipelines, literature workflows, or lab notebooks need accurate sequence or annotation data without leaving the agent environment. The skill documents database coverage and update considerations rather than replacing the gget Python package itself.855installs68Clinical Decision Supportclinical-decision-support is a Claude Code skill from k-dense-ai scientific-agent-skills that produces structured genomic and clinical profile reports for decision-support pipelines. The skill uses a LaTeX article template with tiered evidence color coding across three tiers, plus dedicated sections for mutations, amplifications, and fusions. Reports render in a 10pt letterpaper layout with tabular variant data, booktabs tables, and color-coded tcolorbox callouts for evidence strength. Developers reach for clinical-decision-support when building health-informatics agents that must output clinician-readable genomic profiles with graded evidence rather than raw variant VCF dumps.854installs69Modalmodal is a Python serverless deployment skill from k-dense-ai/scientific-agent-skills that documents the Modal API for grouping, decorating, and executing cloud functions. The reference covers modal.App for deployment units, modal.Function with .remote(), .local(), .spawn(), and .map() for sync, local, async, and parallel execution, plus class-based registration via app.cls(). Developers reach for modal when scientific pipelines, ML inference, or batch ETL jobs need elastic GPU containers instead of fixed VMs or Kubernetes clusters. The skill fits agent-assisted workflows where an AI coder wraps existing Python modules in Modal decorators and validates remote execution paths. It assumes Python familiarity and Modal account setup, and pairs naturally with data-science skills like vaex or geomaster when cloud scale is required.854installs70Vaexvaex is a large-scale tabular data skill from k-dense-ai/scientific-agent-skills for working with datasets too big for in-memory pandas. Vaex DataFrames use lazy evaluation so operations defer until needed, out-of-core processing so data need not fit in RAM, and virtual columns that add computed fields with no memory overhead. The optimized C++ backend targets billion-row-per-second throughput on aggregations and filters. Developers reach for vaex when exploring parquet, HDF5, or CSV files at scales where pandas runs out of memory. The skill covers vaex.open() loading, DataFrame structure, and transformation patterns. It fits data engineers building ETL previews, scientific agents analyzing massive telemetry, or backend pipelines that aggregate huge logs without Spark clusters.854installs71Pufferlibpufferlib is a scientific-agent skill for creating custom reinforcement-learning environments with PufferLib's PufferEnv API and the Ocean suite of 20+ pre-built environments. The skill explains in-place observation, action, and reward buffers, flat observation and action spaces, and native vectorization for both single-agent and multi-agent setups. Developers reach for pufferlib when prototyping RL experiments, porting Gym-style environments, or optimizing environment throughput inside AI coding agents. The guide covers PufferEnv design patterns, performance-oriented in-place operations, and how to select or extend Ocean environments without unnecessary array copies.852installs72Deeptoolsdeeptools is a scientific agent skill that wraps the deepTools CLI for next-generation sequencing analysis, covering bamCoverage for normalized bigWig tracks, bamCompare for treatment-versus-control ratios, multiBamSummary with plotCorrelation heatmaps, and computeMatrix heatmaps around transcription start sites. Example commands include RPGC normalization with effectiveGenomeSize, log2 bamCompare operations, and Pearson correlation plots exported as PNG. Developers reach for deeptools when they need reproducible NGS coverage comparisons and heatmaps from BAM files without manually assembling multi-step shell pipelines from scratch.849installs73Benchling Integrationbenchling-integration is an agent skill for the Benchling REST API v2 at https://{tenant}.benchling.com/api/v2. It documents authentication via API key basic auth or OAuth bearer tokens, stable versus alpha and beta endpoint tiers, and core resources including DNA sequences accessible through curl-ready examples. Developers reach for benchling-integration when building scientific agents that must read or write sequences, lab notebooks, or molecular biology entities without manually browsing Benchling docs each session. The skill maps tenant-scoped base URLs, required HTTP headers, and v2 stability guidelines for production lab automation pipelines.847installs74Pyhealthpyhealth is a clinical machine-learning skill for building healthcare prediction pipelines with PyHealth inside AI coding agents. It covers loading EHR and clinical datasets such as MIMIC-III, MIMIC-IV, eICU, OMOP, SleepEDF, ChestXray14, and EHRShot, then defining tasks like mortality, readmission, length-of-stay, drug recommendation, sleep staging, ICD coding, and EEG events. The skill instantiates models including Transformer, RETAIN, GAMENet, SafeDrug, MICRON, StageNet, AdaCare, and CNN, RNN, or MLP variants, trains with the PyHealth Trainer, computes clinical metrics, and uses ICD, ATC, NDC, and RxNorm code utilities. Developers reach for pyhealth when implementing regulated healthcare ML features rather than generic tabular models.847installs75Scanpyscanpy is a single-cell RNA-seq workflow skill built around a complete Python template using Scanpy, pandas, numpy, and matplotlib. The pipeline loads raw count data from h5ad files, applies QC filtering, normalization, highly variable gene selection, PCA, neighbor graphs, clustering, and cell-type annotation, writing results and figures to configured output directories. Developers reach for scanpy when implementing scRNA-seq analysis in agent-assisted notebooks or scripts and need a reproducible starting structure rather than ad hoc steps. Parameters for input paths, output folders, and QC thresholds are customizable for dataset-specific runs inside Cursor or Claude Code.847installs76Adaptyvadaptyv is an agent integration skill for the Adaptyv Bio Foundry public API at https://foundry-api-public.adaptyvbio.com/api/v1. It documents nine endpoint groups—Experiments, Sequences, Results, Targets, Quotes, Tokens, Updates, and Feedback—so coding agents can create Draft experiments, submit sequences, fetch results, and request quotes with correct request bodies. Developers reach for adaptyv when building scientific agent pipelines that must submit foundry work, poll status, and retrieve structured lab output instead of manually translating OpenAPI into curl or SDK calls. The skill exposes POST /experiments creation, experiment_spec schemas, and related CRUD flows from the bundled endpoint reference.845installs77Deepchemdeepchem is an agent skill providing a comprehensive API reference for DeepChem data handling in scientific ML pipelines. It documents file-format loaders including CSVLoader, UserCSVLoader, SDFLoader for molecular structures, JsonLoader, and ImageLoader, plus biological loaders such as FASTALoader for protein and DNA sequences and FASTQLoader for sequencing reads with quality scores. Developers reach for deepchem when agents must ingest cheminformatics or bioinformatics files, apply DeepChem transformers, and produce train-validation splits for molecular property or toxicity models. The bundled reference organizes core APIs by functionality from raw file ingest through dataset objects ready for model training.845installs78Torchdrugtorchdrug is a molecular machine-learning skill documenting TorchDrug's modular architecture for drug discovery workflows inside AI coding agents. TorchDrug separates representation models in models.py, task definitions in tasks.py, graph data handling in data.py and datasets.py, and shared base utilities in core.py so developers can mix components across experiments. Developers reach for torchdrug when building graph neural network pipelines for molecular property prediction, protein-ligand tasks, or reusable embedding models across multiple objectives. The skill emphasizes swapping representations, reusing datasets, and prototyping task combinations without rewriting core graph logic.845installs79Datamoldatamol is a scientific-agent skill referencing the datamol.conformers module for 3D molecular conformer generation and analysis in Python. It documents dm.conformers.generate with parameters including n_confs, rms_cutoff in Ångströms, minimize_energy with UFF defaults, and embedding methods such as ETKDGv3, plus add_hs and rotatable-bond-aware auto conf counts. Developers reach for datamol when cheminformatics pipelines need diverse low-energy 3D structures, duplicate conformer pruning by RMS threshold, or clustering for docking and ML featurization workflows. The skill fits computational chemistry and drug-discovery agents that must call datamol APIs correctly instead of guessing embedding method names or energy-minimization flags.844installs80Imaging Data CommonsImaging Data Commons is a scientific-agent skill for querying the NCI Imaging Data Commons (IDC) via Google BigQuery when the lighter idc-index tool cannot expose required imaging metadata. The guide is tested against IDC data version v23 and covers prerequisites: a Google account, a billed GCP project (first 1 TB/month free), and google-cloud-bigquery or console access with gcloud application-default login. Use BigQuery when you need all 4000+ DICOM tags instead of idc-index's ~50 tags, segmentation references, structured report measurements, or multi-table joins. Developers reach for this skill during radiology ML cohort assembly, metadata audits, and reproducible SQL pipelines inside agent workflows.844installs81Fluidsimfluidsim is an agent skill for advanced FluidSim turbulence simulation setup in Python. It documents forcing mechanisms—including time-correlated random (tcrandom) forcing with nkmin_forcing, nkmax_forcing, forcing_rate, and tcrandom_time_correlation parameters—and proportional forcing that maintains a target energy distribution. Developers reach for fluidsim when they need sustained turbulence, custom in-script initial conditions, or Fourier-space hooks instead of default solver presets. The skill translates research goals into params.forcing.enable, params.forcing.type, and wavenumber-band injection settings agents can embed directly in simulation scripts. Use it while prototyping HPC or desktop CFD notebooks where forcing type and energy injection rate materially affect flow statistics.843installs82Pydeseq2pydeseq2 is a scientific agent skill that teaches correct PyDESeq2 usage for RNA-seq differential expression analysis in Python. It centers on the DeseqDataSet class, which accepts a samples-by-genes counts DataFrame, sample metadata, and a Wilkinson design formula such as ~condition, then runs dispersion and log fold-change estimation. Developers reach for pydeseq2 when building reproducible DE pipelines and want agents to follow proper normalization, design specification, and LFC fitting instead of improvising pandas-only shortcuts. The bundled API reference covers initialization parameters, core class methods, and the full processing path from count matrices through statistical modeling.843installs83RdkitRDKit is a scientific-agent skill providing a structured Python API reference for the RDKit cheminformatics toolkit inside coding-agent sessions. Coverage centers on rdkit.Chem molecule I/O—MolFromSmiles, MolFromSmarts, MolFromMolFile, MolFromMolBlock, MolFromMol2File, and MolFromMol2Block—with sanitize and removeHs options called out explicitly. Developers reach for RDKit when agents must generate correct parsing code for SMILES strings, SMARTS patterns, MOL/PDB-adjacent structures, or descriptor calculations without hallucinating deprecated calls. The reference is organized by functionality so agents can drop in vetted patterns during HTS triage, ligand preparation, or custom molecular filtering scripts.843installs84AstropyAstropy is a k-dense-ai scientific-agent-skills reference for the astropy.coordinates package and its SkyCoord interface. It shows how to create coordinates from decimal degrees, sexagesimal strings, mixed hour-angle formats, and galactic frames, then transform between systems such as ICRS and galactic. The skill covers array coordinates for batch catalog processing with astropy.units. Use Astropy when agents or developers build Python astronomy apps, cross-match sky catalogs, or convert observation frames in research pipelines. It focuses on coordinate APIs rather than full observatory pipeline orchestration, making it a targeted helper for astropy.coordinates usage in scientific Python workflows.841installs85Cellxgene Censuscellxgene-census is a schema and query reference for the CZ CELLxGENE Census—a versioned single-cell RNA collection built on TileDB-SOMA. The Census organizes data as a SOMACollection with census_info (summary build stats, dataset metadata, stratified cell counts) and census_data organism experiments such as homo_sapiens. Developers and computational biologists use this skill when an agent must interpret available metadata fields, navigate SOMAExperiment structure, and write correct census queries for downstream notebooks or pipelines. It fits early in analysis design when you need to understand what observations, datasets, and summary tables exist before pulling expression matrices. Pair it with Python census APIs when building batch-corrected or cohort-specific single-cell workflows.841installs86Molecular Dynamicsmolecular-dynamics is a scientific agent skill from k-dense-ai/scientific-agent-skills that provides MDAnalysis reference patterns for protein trajectory analysis in Python. The skill covers Universe and AtomGroup construction from topology PDB and trajectory DCD files, atom selection language for protein, backbone, and residue queries, and standard analysis modules including RMSD and RMSF calculations. Developers load trajectories with mda.Universe, inspect n_atoms, n_residues, n_frames, dt, and totaltime attributes, then apply selection strings to isolate structural regions for analysis. Reach for molecular-dynamics when building computational biology pipelines that process MD simulation output without repeatedly consulting MDAnalysis documentation.841installs87ArboretoArboreto is a scientific-agent skill for gene regulatory network (GRN) inference from expression matrices inside terminal-based agent workflows. The skill documents two multiple-regression algorithms—GRNBoost2 (gradient boosting, recommended for large datasets) and GENIE3—that train a model per target gene, rank feature importance, and emit candidate regulators with scores. Developers reach for Arboreto when building single-cell or bulk RNA-seq analysis pipelines that need ranked regulator hypotheses rather than manual feature selection. GRNBoost2 targets tens-of-thousands-of-genes scale where GENIE3 would be slower. The skill fits k-dense-ai/scientific-agent-skills alongside other bioinformatics agent tools.840installs88Esmesm is a reference skill from k-dense-ai/scientific-agent-skills for calling ESMFold2 structure prediction on the Biohub platform via the EvolutionaryScale Python SDK. EvolutionaryScale is migrating hosted inference from Forge to Biohub, and the skill documents API key setup in the Biohub developer console, the ESM_API_KEY environment variable, and esm.sdk.forge client usage against newer Biohub endpoints. Developers reach for esm when building Python AI agents or notebooks that need all-atom protein structure prediction without misconfiguring deprecated Forge URLs or auth. The skill covers install steps, authentication, and client invocation patterns so agents produce valid prediction requests on biohub.ai.840installs89Neurokit2NeuroKit2 is a scientific-agent skill documenting the NeuroKit2 Python library's Bio module for simultaneous physiological signal processing inside coding-agent workflows. The centerpiece is bio_process(), which accepts optional ECG, respiratory (rsp), EDA, EMG, PPG, and EOG arrays with a configurable sampling_rate and returns processed signals plus info metadata in one coordinated call. Developers reach for NeuroKit2 when building wearable, HCI, or clinical research pipelines that must analyze multiple biosignals consistently rather than chaining separate single-modality scripts. The skill wraps signal-specific processors behind unified interfaces suitable for reproducible Python notebooks or batch pipelines driven by an agent.840installs90Zarr Pythonzarr-python is an MIT-licensed skill from k-dense-ai/scientific-agent-skills for storing large N-dimensional arrays with chunking and compression in Python. The skill covers Zarr array creation, parallel I/O patterns, and cloud-native workflows on S3 and GCS using s3fs, requiring Python 3.11 or newer installed via uv pip install zarr. It integrates seamlessly with NumPy, Dask, and Xarray for large-scale scientific computing pipelines that need efficient reads and writes against object storage. Developers reach for zarr-python when building data pipelines that must handle massive multi-dimensional datasets with compressed chunked arrays instead of loading entire arrays into memory.840installs91Dhdna Profilerdhdna-profiler is an advanced profiling skill from k-dense-ai/scientific-agent-skills that scores authored text across multiple cognitive and stylistic dimensions. Domain-specific presets target academic writing with analytical depth, citation patterns, and hedging language; creative writing with metaphor density and narrative arc; and business or executive communication with strategic thinking and social dimensions. Each preset maps focus dimensions such as Linguistic Precision, Domain Expertise, Metacognition, Creative Range, and Intuitive Reasoning to typical topologies like Linear, Dialectic, Spiral, or Web. Developers reach for dhdna-profiler when agents must critique drafts—papers, creative copy, or executive memos—before revision cycles or publication.839installs92Pathmlpathml is a data-management skill from k-dense-ai/scientific-agent-skills for handling large-scale pathology whole-slide images in machine learning pipelines. It documents PathML's HDF5-centric storage for images, masks, features, and metadata with chunked compression, fast random access, and hierarchical organization suited to arbitrarily large slide libraries. Developers reach for pathml when building computational pathology agents that must tile whole-slide images, persist processed subsets, and feed batch loaders without reloading multi-gigabyte WSIs each epoch. The skill covers storage schemas, tile management strategies, and optimized batch processing workflows for downstream model training and analysis.839installs93PydicomPydicom is a k-dense-ai scientific-agent-skills reference for commonly used DICOM tags organized by clinical category such as patient information, study metadata, and image parameters. It documents tag tuples like (0010,0010) PatientName and shows pydicom access via attribute notation ds.PatientName or tuple notation ds[0x0010, 0x0010]. Use Pydicom when agents or developers parse DICOM files, validate metadata fields, or wire hospital imaging ETL without memorizing the standard tag dictionary. The skill accelerates pydicom script authoring for research and healthcare imaging workflows that need correct field names, VR types, and tag lookups during pipeline development.839installs94Depmapdepmap is a Claude agent skill from k-dense-ai/scientific-agent-skills that teaches agents how to interpret Chronos v5+ gene dependency scores from DepMap CRISPR screen data. The guide defines score bands from growth-promoting (>0) through non-essential (0 to −0.3), mild dependency (−0.3 to −0.5), significant dependency (−0.5 to −1.0), and strong essentiality (< −1.0, commonly near −1.0). Chronos corrects copy-number bias, guide RNA efficiency variation, and cell-line growth-rate effects that skew raw CRISPR hits. Bioinformatics engineers and computational biologists invoke depmap inside agentic workflows to flag cancer-selective dependencies, filter common essential genes, and prioritize therapeutic targets from DepMap tables rather than misreading noise as biology.838installs95PymatgenPymatgen is a k-dense-ai scientific-agent-skills reference for pymatgen.analysis modules used in computational materials science. It covers PhaseDiagram construction from ComputedEntry lists, stable entry extraction, energy above hull calculations, decomposition products, and PDPlotter visualization patterns. The skill extends to thermodynamic and electrochemical analyses such as Pourbaix diagrams for corrosion and stability studies. Use Pymatgen when agents or developers automate materials characterization, DFT post-processing, or stability screening in Python research pipelines. It targets developers integrating pymatgen APIs rather than teaching general chemistry, offering concrete module imports and analysis workflows for production scientific code.838installs96Pysampysam is a genomic file toolkit skill from k-dense-ai/scientific-agent-skills covering the Python pysam module built on htslib. It documents read/write access to SAM, BAM, and CRAM alignment files, VCF and BCF variant calls, and FASTA/FASTQ sequences, plus tabix-indexed region queries, pileup coverage analysis, and samtools/bcftools command execution from Python. Developers reach for pysam when building bioinformatics agents or NGS pipelines that must parse alignments, extract genomic regions, calculate coverage, or transform variant data without shelling out manually. The skill is MIT-licensed reference material authored by K-Dense Inc. for sequencing data processing workflows.838installs97Scvi Toolsscvi-tools is an agent guidance skill for probabilistic differential expression (DE) testing in single-cell RNA and related modalities. scvi-tools leverages learned generative models to estimate expression differences between groups with batch-corrected representations, uncertainty quantification on effect sizes, and explicit zero-inflation handling for dropout-heavy sparse counts. Developers reach for this skill when traditional DE methods fail across batches or when zeros dominate the count matrix. The workflow supports flexible comparisons between arbitrary groups or cell types after model training. Use it while building reproducible notebooks or pipelines that need principled Bayesian effect estimates instead of naive fold-change tests on raw counts.838installs98Polars Biopolars-bio is version 1.0 agent skill from K-Dense scientific-agent-skills for the polars-bio Python library, requiring Python 3.11 through 3.14. The library delivers genomic interval operations—overlap, nearest, merge, coverage, complement, and subtract—on Polars DataFrames with streaming and cloud-native I/O for BED, VCF, BAM, and GFF files via s3://, gs://, and az:// URIs using standard cloud SDK environment variables. Bioinformatics engineers reach for polars-bio when bioframe-style interval algebra must run faster at scale. Install with uv pip install polars-bio. The skill fits pipeline builders processing large genomics datasets who already standardize on Polars instead of pandas.835installs99Treatment Planstreatment-plans is an agent skill from K-Dense scientific-agent-skills that produces concise 3-4 page medical treatment plans rendered in LaTeX and exported to PDF for clinical specialties including general medicine, rehabilitation, mental health, chronic disease, perioperative care, and pain management. Plans follow SMART goal frameworks with evidence-based interventions, minimal inline citations, and HIPAA-aware professional formatting prioritizing brevity and clinical actionability. The skill uses LLM-powered steps requiring an OPENROUTER_API_KEY environment variable. Clinical software developers and health informatics engineers reach for treatment-plans when automating structured care plan documents—not general Markdown notes. Allowed tools include Read, Write, Edit, and Bash for LaTeX compilation.835installs100Diffdockdiffdock is a scientific agent skill from k-dense-ai/scientific-agent-skills that wraps DiffDock inference for protein-ligand docking inside coding-agent sessions. Inputs accept CSV rows with complex_name, protein_path, ligand_description as SMILES strings, or inline protein_sequence when structures are sequence-only, plus ligand SDF paths. A custom inference configuration template sets model_dir, confidence_model_dir, and checkpoint paths under workdir/v1.1 for score and confidence models. Developers reach for diffdock when cheminformatics or structural biology workflows need pose predictions from conversational prompts or tabular batch files without manually scripting separate CLI pipelines.834installs101FlowioFlowIO is an agent skill around the lightweight FlowIO Python library (skill metadata version 1.0, BSD-3-Clause) for reading and writing Flow Cytometry Standard files. It parses FCS versions 2.0, 3.0, and 3.1, extracts event data as NumPy arrays, reads metadata and channel definitions, and converts outputs to pandas DataFrames or CSV for downstream pipelines. Developers reach for FlowIO when backend services or ETL jobs need minimal-dependency cytometry file ingestion without heavier flow-cytometry suites. The skill fits basic file operations, metadata inspection, and event extraction in scientific data workflows.834installs102Genimlgeniml is a skill from k-dense-ai/scientific-agent-skills for BEDspace, which applies the StarSpace model to genomic region sets and their metadata labels in one low-dimensional embedding space. The workflow spans four sequential operations starting with preprocess, enabling cross-modal queries such as finding regions similar to a cell-type or tissue label. Developers reach for geniml when building agents or pipelines that must search genomic regions with metadata-aware similarity, joint analysis of BED intervals and experimental conditions, or unified vector retrieval across regions and labels. It complements single-modal genomic indexes by training coupled region-and-label representations.834installs103Medchemmedchem is a skill from k-dense-ai/scientific-agent-skills referencing medchem 2.0.5 APIs for filtering molecules with RuleFilters from medchem.rules. The RuleFilters class accepts a rule_list of named or callable rules, runs across sequences of SMILES or RDKit Mol objects with optional parallel n_jobs, and returns a pandas DataFrame with mol and pass_all columns plus per-rule results. Developers reach for medchem when agents must enforce medicinal chemistry constraints—lead-likeness, structural alerts, or custom filters—inside computational chemistry or drug-discovery pipelines. Official API docs live at medchem-docs.datamol.io for deeper rule configuration.834installs104PhylogeneticsPhylogenetics is a scientific-agent skill (version 1.0) that reconstructs evolutionary history from genes, proteins, or genomes via a standard pipeline: MAFFT for multiple sequence alignment, IQ-TREE 2 for maximum-likelihood tree inference, FastTree for fast NJ/ML alternatives, and ETE3 or FigTree for visualization. Developers invoke it for microbial genomics, viral phylodynamics, protein family analysis, and molecular clock studies when they need agent-guided command sequences instead of ad hoc shell scripts. The skill encodes when to pick IQ-TREE 2 versus FastTree and how to move from raw FASTA inputs to annotated tree figures suitable for publication or downstream comparative analysis.834installs105Scveloscvelo is a skill from k-dense-ai/scientific-agent-skills (version 1.0, BSD-3-Clause) for RNA velocity analysis with scVelo on single-cell RNA-seq data. scVelo models mRNA splicing kinetics using unspliced pre-mRNA and spliced mature mRNA abundances to infer cell state transition directions, compute latent time, and identify driver genes. The skill complements Scanpy and scVI-tools for trajectory inference beyond static clustering. Developers reach for scvelo inside Claude Code or Cursor when analyzing differentiation dynamics, pseudotime directionality, or velocity-based gene regulation hypotheses on scRNA-seq datasets that include splicing-aware counts.834installs106Pennylanepennylane is version 1.1 agent skill from K-Dense scientific-agent-skills for the hardware-agnostic PennyLane quantum machine learning framework. The skill guides developers training quantum circuits via gradients, building hybrid quantum-classical models, and running variational algorithms such as VQE and QAOA with PyTorch or JAX backends. PennyLane abstracts device portability across IBM, Google, Rigetti, and IonQ simulators and hardware. Reach for pennylane when the task is differentiable quantum programming—not IBM-only Qiskit optimizations or open-system modeling with QuTiP. The skill covers quantum neural networks and gradient-based circuit training for research prototypes and ML pipelines.833installs107CirqCirq is a scientific-agent skill (version 1.0, Apache-2.0) for Google Quantum AI's open-source Python framework used to design, simulate, and execute quantum circuits on Google quantum computers and simulators. The skill steers coding agents toward noise-aware circuit design, quantum characterization experiments, and low-level gate-level optimization rather than general-purpose scientific Python. Developers reach for Cirq when targeting Google Quantum AI hardware, modeling decoherence, or comparing against Qiskit for IBM, PennyLane for quantum ML autodiff, or QuTiP for physics simulations. Allowed tools include Read, Write, Edit, and Bash so agents can scaffold circuits, run local simulators, and prepare jobs for Google backends from Cursor or Claude Code.832installs108GtarsGtars is a scientific-agent skill (version 1.0) for Genomic Tools and Algorithms in Rust, a high-performance toolkit with Python bindings for manipulating genomic interval data in computational genomics and machine learning. The skill helps developers run overlap detection, coverage analysis, fragment analysis, reference sequence management, and tokenization pipelines on BED files and coverage tracks without hand-rolling slow Python loops. Agents use it when building genomics ETL, peak-calling post-processing, or preparing interval features for ML models. Allowed tooling focuses on installing the library and generating correct Rust-backed Python calls for large-scale interval operations that standard pandas workflows struggle to keep fast.832installs109Labarchive Integrationlabarchive-integration is version 1.0 agent skill from K-Dense scientific-agent-skills that enables programmatic access to LabArchives electronic lab notebooks through its REST API. Researchers and pipeline engineers use it to read and write notebook entries, manage attachments, generate reports, back up notebooks, and connect workflows with Protocols.io, Jupyter, and REDCap. The skill fits developers automating ELN data flows instead of manual browser uploads. Reach for labarchive-integration when an agent or script must create, query, or sync LabArchives notebook records as part of a reproducible research pipeline.832installs110MatchmsMatchms is an agent skill for the open-source matchms Python library (skill metadata version 1.0, Apache-2.0) focused on mass spectrometry data processing and metabolite identification. It imports spectra from multiple formats, standardizes metadata, filters peaks, calculates spectral similarities including cosine and modified cosine, and searches spectral libraries to identify unknown compounds. Developers reach for Matchms in metabolomics agent pipelines when comparing experimental spectra against reference libraries. The skill notes that full LC-MS/MS proteomics pipelines should use pyopenms instead of matchms.832installs111PrimekgPrimeKG is an agent skill for the Precision Medicine Knowledge Graph originally from Harvard MIMS (skill metadata version 1.0) that unifies over 20 primary databases and scientific literature into one biomedical graph. The graph contains over 100,000 nodes and 4 million edges across 29 relationship types, including drug-target, disease-gene, and phenotype-disease associations. Developers reach for PrimeKG when agent or backend workflows need to search genes, proteins, drugs, diseases, and phenotypes or traverse multiscale biological relationships during precision-medicine research. Key capabilities include node search, relationship lookup, and structured queries across integrated biomedical entities.832installs112TiledbvcfTileDB-VCF is a scientific-agent skill (version 1.0, MIT license) for a high-performance C++ library with Python and CLI interfaces that stores variant-call data on TileDB sparse arrays. The skill guides scalable VCF/BCF ingestion, incremental sample addition without expensive merges, compressed storage, parallel variant queries, and export for population genomics workloads. Developers reach for it when cohort VCF files outgrow flat-file grep workflows and need a queryable, append-friendly variant store. Agents produce correct TileDB-VCF CLI commands and Python API usage for ingestion pipelines, filtered exports, and incremental cohort updates from Cursor or Claude Code.832installs113Etetoolkitetetoolkit is a k-dense-ai/scientific-agent-skills module (version 1.0, GPL-3.0) wrapping ETE—the Environment for Tree Exploration—for phylogenomic research. The skill supports Newick and NHX tree loading, hierarchical manipulation, evolutionary event detection, orthology and paralogy analysis, NCBI taxonomy integration, and visualization export to PDF and SVG. Developers reach for etetoolkit when natural-language requests need phylogenetic tree operations, clustering analysis, or publication-ready tree figures without manually scripting each ETE API call.831installs114HistolabHistolab is an agent skill for the Histolab Python library (skill metadata version 1.1, Apache-2.0) that processes whole slide images through tissue detection, tile extraction, and stain normalization for H&E images. It requires Python 3.8–3.11 with histolab 0.7.0, OpenSlide system libraries on Linux or macOS, and optional pooch for sample data. Developers reach for Histolab when preparing tile datasets, running basic slide preprocessing, or standing up simple digital-pathology pipelines without the heavier spatial-proteomics stack. The skill notes that advanced multiplexed imaging or deep-learning spatial workflows should use pathml instead.831installs115MolfeatMolfeat is an agent skill for the molfeat Python featurization hub (skill metadata version 1.0, Apache-2.0) that unifies 100+ pre-trained embeddings and hand-crafted featurizers for molecular machine learning. It converts SMILES and chemical structures into ECFP fingerprints, MACCS keys, molecular descriptors, and pretrained transformer embeddings such as ChemBERTa for QSAR and virtual screening workflows. The skill requires Python 3.9–3.10 because molfeat 0.11.0 does not support Python 3.11+, plus datamol, PyTorch, and optional extras for GNN or transformer models. Developers reach for Molfeat when standardizing molecular feature extraction across multiple model backends in cheminformatics pipelines.831installs116Pyopenmspyopenms is a skill from k-dense-ai/scientific-agent-skills covering PyOpenMS core data structures for LC-MS experiments. It shows creating an MSExperiment, loading mzML files with MzMLFile().load, and inspecting spectrum counts, chromatogram counts, retention-time ranges, and individual spectra. PyOpenMS exposes C++ mass-spec objects through Python bindings, so the skill emphasizes correct object usage for proteomics and metabolomics pipelines. Developers reach for pyopenms when agents must read mzML data, summarize experiment metadata, or manipulate spectra and chromatograms inside Claude, Cursor, or Codex workflows without leaving the coding session.831installs117Usfiscaldatausfiscaldata is a version 1.1 MIT K-Dense skill for the free U.S. Department of the Treasury Fiscal Data REST API at api.fiscaldata.treasury.gov. The skill guides agents to fetch national debt (Debt to the Penny), Daily and Monthly Treasury Statements, securities auctions, interest rates, foreign exchange rates, savings bonds, and revenue or spending statistics with no API key or registration. Developers reach for usfiscaldata when embedding live U.S. federal finance data into dashboards, research scripts, or agent tools that must stay keyless and publicly accessible. Allowed tools include Read, Write, Edit, and Bash for fetching and shaping API responses in code.831installs118Protocolsio IntegrationProtocols.io Integration is an agent skill (metadata version 1.0) that connects AI coding agents to the protocols.io API for end-to-end scientific protocol management. It covers searching and publishing protocols, editing steps and materials, handling discussions and comments, organizing workspaces, and uploading files within lab documentation workflows. Developers reach for Protocols.io Integration when building experiment-tracking tools, collaborative protocol editors, or agent-driven lab assistants that must read and write protocols.io resources programmatically. The skill applies to protocol discovery, collaborative protocol development, experiment tracking, and scientific documentation automation.830installs119PytdcPyTDC is a scientific-agent skill (version 1.0, MIT license) for Therapeutics Data Commons, an open-science platform of AI-ready drug discovery datasets and benchmarks spanning the therapeutics pipeline. The skill helps developers load standardized ADME, toxicity, drug-target interaction, and molecular oracle datasets with meaningful scaffold splits and evaluation metrics for single-instance and multi-instance prediction tasks. Reach for PyTDC when prototyping pharmacological predictors, benchmarking therapeutic ML models, or avoiding one-off dataset wrangling in computational drug discovery. Agents guided by this skill generate correct PyTDC API calls, pick appropriate benchmark splits, and wire datasets into training scripts inside Cursor or Claude Code.830installs120Opentrons IntegrationOpentrons Integration is a version 1.0 K-Dense scientific-agent skill for writing Opentrons Protocol API v2 Python that runs on OT-2 and Flex robots. The skill guides agents through pipetting steps, labware placement, heater-shaker and thermocycler module control, and production protocol structure aligned with official Opentrons hardware APIs. Developers reach for Opentrons Integration when automating bench protocols in Python and needing official API compatibility instead of multi-vendor abstractions. The readme directs broader multi-equipment automation to pylabrobot, keeping this skill focused on Opentrons-native production protocols.829installs121Neuropixels Analysisneuropixels-analysis is a K-Dense Inc. MIT-licensed skill (version 2.1) for end-to-end Neuropixels extracellular recording analysis using SpikeInterface inside Claude Code or Cursor. The pipeline covers loading SpikeGLX, Open Ephys, and NWB data, preprocessing, drift and motion correction, Kilosort4 and CPU spike sorting, quality metrics, and unit curation including threshold-based, UnitRefine model-based, and AI-assisted visual review. Developers reach for neuropixels-analysis when processing Neuropixels 1.0 or 2.0 recordings rather than building SpikeInterface workflows manually. Optional Claude API features require ANTHROPIC_API_KEY for AI-assisted curation steps.828installs122QutipQuTiP is a version 1.0 K-Dense scientific-agent skill built around the BSD-3-Clause Quantum Toolbox in Python for simulating closed and open quantum systems. The skill helps agents set up master equations, Lindblad dissipators, cavity QED models, and decoherence studies directly inside Claude Code or Cursor workflows. Developers reach for QuTiP when researching open-system dynamics, quantum optics, or educational physics simulations in Python. The readme explicitly excludes circuit-based quantum computing and points those workloads to Qiskit, Cirq, or PennyLane for algorithm and hardware execution paths.827installs123RowanRowan is a K-Dense scientific-agent skill for Rowan's cloud-native molecular modeling platform accessed via Python 3.12+ and a ROWAN_API_KEY. The skill covers pKa and macropKa prediction, conformer and tautomer ensembles, docking and analogue docking, protein-ligand cofolding, MSA generation, molecular dynamics, permeability, and descriptor batch pipelines programmatically. Developers reach for Rowan when screening small molecules or proteins in multi-step chemistry workflows that would otherwise require maintaining local GPU clusters. The platform is proprietary and API-key gated, optimized for programmatic batch screening rather than one-off desktop modeling GUIs.827installs124Cobrapycobrapy is a K-Dense scientific agent skill (version 1.1, GPL-2.0) for COBRApy constraint-based metabolic modeling, covering FBA, FVA, gene knockouts, flux sampling, and SBML model analysis for systems biology and metabolic engineering. The skill requires Python 3.9+ because cobra 0.30+ dropped Python 3.8, installs via `uv pip install`, and defaults to the GLPK solver through swiglpk with optional CPLEX or Gurobi. Model loading can fetch from bundled data, BiGG, or BioModels when network access is available. Developers reach for cobrapy when agent sessions must run flux balance analysis, knockout studies, or sampling on genome-scale metabolic models without hand-writing boilerplate solver setup and model import code.826installs125Dnanexus Integrationdnanexus-integration is an agent skill from k-dense-ai/scientific-agent-skills (version 1.1, K-Dense Inc.) that guides DNAnexus cloud genomics development using the dxpy Python SDK. The skill covers building apps and applets, uploading and downloading biomedical data, running workflows, and working with FASTQ, BAM, and VCF formats, with optional DX_SECURITY_CONTEXT from dx login and DX_ASSET_BWA for BWA examples. Developers reach for dnanexus-integration when an AI agent must authenticate to DNAnexus, scaffold genomics pipelines, or execute analyses on managed cloud datasets.826installs126Lamindblamindb is a K-Dense scientific agent skill (version 1.1, Apache-2.0) for LaminDB, an open-source lineage-native lakehouse for biological datasets and models. The skill covers setup, artifact registration, query and search, lineage tracking, validation, ontology-backed annotation with Bionty, collections, branches, storage across local filesystems and S3, and workflow integrations. Developers reach for lamindb when they need FAIR-compliant, reproducible biology data management inside Claude Code or Cursor instead of ad hoc file copies and untracked notebooks. It guides agents through registering artifacts, tracing provenance, and validating datasets against ontologies so downstream analysis stays queryable and auditable.826installs127Pylabrobotpylabrobot is a K-Dense scientific agent skill (version 1.0, MIT license) for PyLabRobot, a hardware-agnostic pure Python SDK for automated laboratories. The skill guides agents when controlling Hamilton, Tecan, Opentrons, plate readers, pumps, heater shakers, and incubators through one programming model, including simulation for complex multi-vendor workflows. Developers reach for pylabrobot when lab scripts must span equipment from different vendors or need unified abstractions instead of vendor-specific protocol APIs. The skill notes that Opentrons-only projects may be simpler with a dedicated Opentrons integration, but multi-vendor and simulation-heavy setups benefit from PyLabRobot's unified interface.826installs128Qiskitqiskit is a k-dense-ai/scientific-agent-skills module (version 1.0, Apache-2.0) for IBM's open-source quantum computing framework, cited as the world's most popular with 13M+ downloads. The skill targets Qiskit Runtime for production workloads, IBM hardware execution, quantum error mitigation, and circuit optimization on simulators or real quantum processors. README guidance routes Google hardware to cirq, gradient-based quantum ML to pennylane, and open quantum systems to qutip. Developers reach for qiskit when agent-assisted workflows need IBM Quantum deployment rather than alternative quantum SDKs.826installs129Latchbio Integrationlatchbio-integration is an agent skill from k-dense-ai/scientific-agent-skills (version 1.0, K-Dense Inc.) for the Latch Python framework that deploys bioinformatics workflows as serverless pipelines on Flyte. The skill covers @workflow and @task decorators, LatchFile and LatchDir cloud data types, resource configuration, and integration of Nextflow and Snakemake pipelines into Latch deployments. Developers reach for latchbio-integration when an AI agent must scaffold reproducible bioinformatics workflows, package scientific computing tasks, and publish serverless pipeline definitions rather than run one-off local scripts.825installs130Research Lookupresearch-lookup is a MIT-licensed skill from k-dense-ai/scientific-agent-skills that automatically routes research queries to the best backend available. parallel-cli provides the primary fast web search path; the Parallel Chat API handles deep research when PARALLEL_API_KEY is set; and Perplexity sonar-pro-search via openrouter.ai covers academic paper searches when OPENROUTER_API_KEY is configured. Query text is transmitted to api.parallel.ai and, for academic searches, to openrouter.ai, so developers must supply the appropriate API keys. The skill supports finding papers, gathering research data, and verifying scientific information inside agent sessions with Read, Write, Edit, and Bash tool access. Developers reach for research-lookup during literature reviews, hypothesis checks, or when agent workflows need cited, up-to-date scientific sources.825installs131Omero Integrationomero-integration is a microscopy data management skill from k-dense-ai/scientific-agent-skills that connects Claude, Cursor, or Codex agents directly to an OMERO server using Python. The skill retrieves image datasets, analyzes pixels, manages ROIs and annotations, and runs batch processing pipelines for high-content screening. Required environment variables include OMERO_HOST, OMERO_USER, and OMERO_PASSWORD, with OMERO_PORT defaulting to 4064. Developers reach for omero-integration when building automated microscopy analysis pipelines, pulling plate-level screening data, or managing ROI annotations without writing OMERO client boilerplate from scratch.824installs132Glycoengineeringglycoengineering is a scientific analysis skill from k-dense-ai/scientific-agent-skills (version 1.0, author Kuan-lin Huang) that scans protein sequences for N-glycosylation sequons (N-X-S/T), predicts O-glycosylation hotspots, and routes workflows to curated tools including NetOGlyc, GlycoShield, and GlycoWorkbench. Glycosylation is the most common post-translational modification, affecting over 50% of all human proteins and influencing folding, stability, immune recognition, and pharmacokinetics. Developers reach for glycoengineering when optimizing therapeutic antibodies, engineering glycoprotein stability, or designing vaccines that depend on glycan shielding and immune epitope control.823installs133Ginkgo Cloud Labginkgo-cloud-lab is a scientific agent skill (version 1.0) for submitting and managing protocols on Ginkgo Bioworks Cloud Lab at cloud.ginkgo.bio. The platform executes protocols on Reconfigurable Automation Carts (RACs) in Ginkgo's autonomous lab infrastructure. Developers and researchers use the skill for cell-free protein expression validation or optimization, fluorescent pixel art generation, and end-to-end ordering workflows including protocol selection, input preparation, pricing review, and submission. Reach for ginkgo-cloud-lab when an agent must interact with Ginkgo Cloud Lab services instead of manual web portal steps, especially for repeatable biological protocol runs requiring remote hardware.819installs134Hugging Sciencehugging-science is a K-Dense Inc. skill (metadata version 1.1) that indexes scientific resources for AI and ML work across biology, chemistry, physics, astronomy, climate, genomics, materials, medicine, ecology, energy, engineering, math, drug discovery, protein design, weather modeling, theorem proving, single-cell analysis, and PDE solving. The curated, LLM-friendly catalog helps agents discover datasets, models, blog posts, and interactive Spaces, then use them via Hugging Face datasets, transformers, the HF Inference API, gradio_client, and methodology citations. Reach for hugging-science when scientific domain questions need vetted HF assets instead of generic web search. It suits ML engineers and research coders building biology or chemistry agents who want structured discovery paths and reproducible citations rather than ad-hoc model hunting.751installs135Autoskillautoskill is a k-dense-ai/scientific-agent-skills MIT-licensed meta skill that observes the developer's screen through the screenpipe daemon on local port 3030, detects repeated research workflows, matches them against existing scientific-agent-skills, and drafts new skills or composition recipes for gaps. Detection runs locally; only redacted cluster summaries reach the LLM, and the skill refuses to run if screenpipe is unreachable. Allowed tools include Read, Write, Edit, and Bash. Developers reach for autoskill when they want skills grounded in actual work patterns rather than manually authoring SKILL.md files from scratch.724installs136Exa Searchexa-search is a k-dense-ai scientific-agent-skills toolkit for developers who need higher-quality web research than generic search. Built on the exa-py Python SDK with an EXA_API_KEY, it supports semantic lookups, optional `category=research paper` filtering, academic domain filters, and batch URL extraction for pages, articles, and PDFs. Use exa-search when an agent must search the web, fetch a page, or extract scholarly content with MIT-licensed tooling and internet access. Triggers include search, look up, fetch, and extract requests where technical accuracy matters. Reach for exa-search during literature reviews, benchmark sourcing, API documentation discovery, or verifying claims against current papers instead of relying on undifferentiated web search alone.687installs137Bidsbids is version 1.0 of a scientific agent skill in k-dense-ai/scientific-agent-skills for working with Brain Imaging Data Structure datasets across 11 modalities including MRI, EEG, MEG, iEEG, PET, microscopy, NIRS, motion capture, EMG, MR spectroscopy, and behavioral data. It guides agents through BIDS directory layout, dataset_description.json setup, JSON sidecar metadata, .bidsignore configuration, pybids queries by subject/session/task entities, and bids-validator compliance checks before OpenNeuro or DANDI upload. Conversion workflows cover HeuDiConv, dcm2bids, and BIDScoin for DICOM-to-BIDS transforms, plus derivative creation for preprocessed outputs. The skill bundles reference materials including bids_schema.json, beps.yml listing BEP extension proposals, and conversion tool guides. Neuroimaging engineers and computational researchers reach for bids when standardizing raw scanner exports into shareable, validator-clean BIDS trees.542installs138Pacsomaticpacsomatic (version 1.0) in k-dense-ai/scientific-agent-skills is an operator toolkit for nf-core/pacsomatic matched tumor-normal analysis from BAM files. A single helper entrypoint, scripts/run_pacsomatic.py, validates patient and sample identifiers, checks BAM and reference paths, writes pacsomatic-compliant samplesheets (patient,sample,status,bam,pbi), generates params YAML and launch scripts, and supports dry-run validation or live execution locally or on schedulers. Supported executors include LSF, Slurm, PBS, and SGE with resource flags for CPUs, memory, walltime, and queue. Developers reach for pacsomatic when preparing somatic mutation runs, fixing samplesheet errors, submitting Slurm jobs, or triaging .nextflow.log failures without hand-assembling nextflow run commands. The skill enforces routing through the helper script and returns command paths, validation status, and scheduler job IDs per its agent response contract.485installs139Liteparseliteparse skill equips Claude to parse scientific PDFs for agents: extract sections, citations, tables, and metadata into structured chunks suitable for literature review, method extraction, and grounded question answering in research automation workflows.445installs140Bulk RnaseqGuides bulk RNA-seq analysis from FASTQ through QC, alignment or pseudoalignment, count aggregation, normalization, and statistical testing for differential expression in transcriptomics research pipelines.410installs141Pathway EnrichmentPathway-enrichment connects scientific agents to pathway databases so gene lists and biomarkers gain curated ontology IDs, hierarchical context, and cross-database mappings for downstream interpretation and reporting.409installs142Nextflownextflow skill guides Claude to build reproducible scientific workflows: write Nextflow DSL pipelines, wire container images, configure cloud or HPC executors, manage params and resumes, and orchestrate multi-step bioinformatics analyses with provenance-friendly execution.407installs143Experimental Designexperimental-design is a Claude Code skill for design & ui/ux. It helps solo builders move faster with AI-assisted development.310installs144Statistical Powerstatistical-power is a Claude Code skill for ai & agent building. It helps solo builders move faster with AI-assisted development.298installs145Pi Agentpi-agent is a Claude Code skill for ai & agent building. It helps solo builders move faster with AI-assisted development.288installs146Arborarbor is a Claude Code skill for ai & agent building. It helps solo builders move faster with AI-assisted development.272installs