k-dense-ai/scientific-agent-skills
99 skills55k installs3M starsGitHub
Install
npx skills add https://github.com/k-dense-ai/scientific-agent-skillsSkills in this repo
1Scientific WritingScientific Writing is an agent skill that teaches solo builders and small teams how to apply the scientific_report.sty LaTeX package for polished scientific reports, technical write-ups, and white papers. It is aimed at indie researchers, ML engineers, and founders who need citable, conference-ready PDFs without hiring a typesetter. Use it when you are drafting methods, results, or architecture narratives that must look credible to investors, reviewers, or enterprise buyers. The guide covers the color system across blues and greens, when to use each box environment, table styling, and statistic notation commands so agents do not invent ad-hoc LaTeX macros. It matters because inconsistent formatting undermines trust in otherwise solid work, and a single shared style speeds iteration from notebook exports to final deliverables.690installs2Scientific VisualizationScientific-visualization is an agent skill that packages colorblind-friendly color palettes and matplotlib-oriented utilities for scientific and technical plotting. Solo and indie builders use it when building dashboards, research write-ups, or internal analytics views where default rainbow or red-green schemes exclude readers or fail review. The skill centers on widely cited palette families—Okabe-Ito, Wong, and Paul Tol variants—plus lists suited to categorical series and notes on sequential colormaps for continuous data. Rather than ad-hoc color picking in chat, the agent applies consistent, citable palette choices and documents how to wire them into plots. It fits builders who ship Python-backed visualizations and want accessibility baked into figures before launch or publication, without a separate design pass.671installs3Scientific Critical ThinkingScientific Critical Thinking is a journey-wide agent skill that equips solo builders and technical founders to spot systematic errors in reasoning, experimental design, and evidence synthesis. Whether you are validating a startup idea with user studies, interpreting A/B tests, reading ML papers for a feature, or drafting a methods section, the skill frames common biases—confirmation bias, hindsight bias, publication bias—and pairs each with practical mitigations like preregistration, seeking disconfirming evidence, and publishing null results. It is procedural knowledge rather than a single automation: the agent applies the framework when you analyze claims, design studies, or review literature so you do not overfit narratives to noisy data. Indie teams without a dedicated research ops role benefit from having bias names and guardrails on tap during Validate and Ship reviews. It does not run experiments or collect data for you; pair it with your stats stack and domain datasets.670installs4Scientific Brainstormingscientific-brainstorming is a journey-wide agent skill that gives solo researchers and indie technical founders structured frameworks—starting with SCAMPER—for generating and refining scientific ideas before locking a protocol or build plan. The reference walks through seven transformation lenses with domain-specific prompts: substituting materials or models, combining omics or field and lab data, adapting solutions from other disciplines, and similar moves for modify, repurpose, eliminate, and reverse. Agents consult it when a scientist asks for a named methodology or when vanilla brainstorming is too shallow for experimental design tradeoffs. Because triggers span early research through method improvement, it stays useful whenever you need disciplined creativity—not only at project kickoff.665installs5Literature Reviewliterature-review is an agent skill packaged as a rigorous academic template for solo researchers, indie R&D builders, and technical founders who need citable synthesis before building. It guides you through title metadata, abstract, introduction with background and significance, and a methodology section that names databases, selection criteria, and quality assessment expectations. Whether you run a narrative scan or aim at systematic review standards, the scaffold reminds you to state PROSPERO or OSF registration, PRISMA compliance, and explicit primary research questions. For agentic workflows, the skill is a template pattern: the agent fills sections from your sources rather than inventing a one-page summary. Pair it with search and citation tools separately; this skill delivers the document shape reviewers and collaborators expect.652installs6Scientific SchematicsScientific-schematics is an agent skill that encodes best practices for scientific diagrams so solo researchers and technical founders do not guess which export format journals will reject. It prioritizes vector deliverables—PDF for LaTeX and universal print paths, EPS for legacy systems, SVG where online venues allow—and reserves TIFF or PNG for microscopy or photo composites that truly need raster data, always with lossless compression discipline and 300 DPI at final size. The skill also calls out formats to avoid, such as JPEG artifacting on line art, which saves painful resubmission cycles. Accessibility and effective visual communication are treated as first-class requirements alongside file mechanics, which helps indie labs without a dedicated graphics editor. Use it when you are preparing figures for a paper, grant, technical report, or open-science repo README figures—not for marketing hero images. Complexity is beginner-friendly rules with intermediate application when you must merge raster microscopy with vector overlays.643installs7Scientific SlidesScientific-slides provides a production-oriented LaTeX Beamer starter for solo researchers and indie technical founders who need conference-ready decks without rebuilding theme plumbing each time. The template encodes UTF-8 encoding, Madrid/beaver styling, stripped navigation symbols, numbered footers, math and table packages, and authoryear bibliographies suited to scientific talks. Builders use it in the docs phase when translating validation insights or ship milestones into slide narratives—demo days, grant pitches, or paper companion talks. The agent fills frame content, drops figures into ./figures/, and maintains references.bib while preserving the structural comments for outline and institution blocks. It assumes a local LaTeX toolchain rather than Google Slides export, and complements writing-heavy skills by focusing on visual academic presentation format.636installs8MatplotlibMatplotlib is Python’s foundational plotting library for static, animated, and interactive visuals. This agent skill teaches when to reach for low-level control versus higher-level stacks, how to work safely across pyplot and the object-oriented hierarchy, and how to produce figures that survive scientific and product documentation reviews. Solo builders shipping data-heavy SaaS, research tooling, or content backed by metrics use it while coding notebooks, backend analytics, and ship-phase reports. It is the right default when you need novel plot types, pixel-perfect styling, or format-specific export—not when you only need one-line statistical summaries or hosted dashboards. The skill aligns with K-Dense scientific-agent-skills and pairs naturally with pandas-centric pipelines and Jupyter iteration before you freeze assets for launch or growth content.631installs9Citation ManagementCitation Management is an agent skill backed by a structured BibTeX template file showing properly formatted entries for journal articles, books, and related source types. Solo builders and indie researchers use it while documenting experiments, literature reviews, or manuscript drafts so references stay validator-friendly and consistent across projects. The skill is phase-specific to Build documentation: it does not fetch citations from the web but gives canonical field layouts—authors, title, journal, year, volume, pages, DOI—that you or your agent can mirror when adding new keys. Pair it with your own reference manager or search tools when you need live metadata retrieval.627installs10Peer ReviewPeer Review is an agent skill for solo researchers, indie lab leads, and small teams who must produce formal referee reports or structured grant critiques without guessing what editors expect. It walks through systematic assessment of methods, statistical reporting, reproducibility and data availability, ethical conduct, and adherence to common reporting guidelines such as CONSORT, STROBE, and PRISMA. The workflow is optimized for actually writing review text and revision-oriented comments, not for quick gut-checks on a headline claim—that boundary is documented so you pair it with scientific-critical-thinking or scholar-evaluation when you need different lenses. Agents may read, write, edit, and run Bash per skill metadata, so you can iterate review drafts in your repo alongside the manuscript. Prism lists it for builders who ship research artifacts, preprints, or grant packages and want repeatable rigor before submission deadlines.625installs11PdfThe pdf skill teaches coding agents to complete PDF forms without guessing field geometry or widget types. Solo and indie builders use it when contracts, grant applications, tax packets, or vendor onboarding PDFs must be filled consistently inside Claude Code, Cursor, or similar environments. The SKILL.md enforces a hard sequence: never jump to implementation until fillability is known. Agents run python scripts/check_fillable_fields, then branch to fillable-field instructions (extract_form_field_info.py producing structured JSON) or non-fillable-field workflows described later in the skill. That split reduces silent corruption of flat PDFs and mis-typed AcroForm values. The approach is procedural and filesystem-local—ideal when you already have the PDF on disk and want repeatable agent runs rather than a separate SaaS form product. It pairs well with compliance and launch paperwork, but stays a task workflow rather than a journey-wide methodology.623installs12Scikit LearnScikit-learn is a reference-oriented agent skill that helps solo builders implement rigorous model selection without memorizing every sklearn.model_selection API. It documents train-test splits—including stratified and three-way splits—plus cross-validation strategies such as KFold and StratifiedKFold for imbalanced classification. The material is structured for agents generating evaluation notebooks, batch training scripts, or feature pipelines where leakage and split hygiene matter. Use it when you are comparing models, tuning hyperparameters, or standardizing validation methodology in a Python ML slice of your product. It assumes scikit-learn as the stack anchor and pairs well with experimentation or deployment skills once you lock a model. Keep evaluation metrics and business constraints in your own spec; this skill supplies the procedural sklearn patterns.621installs13Markdown Mermaid WritingMarkdown Mermaid Writing is a scientific documentation skill from k-dense-ai that teaches agents how to produce rigorous markdown reports where Mermaid diagrams are the authoritative diagram source—not screenshots or separate draw.io exports. Solo builders and researchers use it when they need reproducible writeups (lab workflows, method pipelines, analysis narratives) that still read well in GitHub, Notion, or static sites. The bundled example walks through a CRISPR gene-editing efficiency study: overview with quantified findings, a five-stage flowchart with GC content, yield, and viability checkpoints, and NGS validation staging. Agents learn to pair narrative claims with structured diagram metadata for accessibility. It suits anyone shipping agent-generated research memos, experiment logs, or architecture decision records that must include flow, sequence, or state diagrams without leaving the markdown file.613installs14Statistical AnalysisStatistical Analysis is an agent skill from scientific-agent workflows that teaches solo builders and small research teams how to validate assumptions before trusting inferential results. It emphasizes documenting checks with both plots and formal tests, understanding when common tests are robust (such as larger per-group samples for normality), and choosing remedial models when independence or distribution assumptions fail. The content spans study-design reasoning through autocorrelation and clustered-data concerns, so it supports experiment readouts during validation as well as ongoing analytics during growth. It is aimed at agents helping write analysis reports, not at replacing a statistician for regulated trials. Use it when you have output from t-tests, ANOVA, regression, or correlation and need a disciplined checklist before you ship conclusions to stakeholders.611installs15Pptxpptx is an agent skill for solo and indie builders who need to edit real PowerPoint files without living inside the desktop app. It teaches a template-first workflow: inspect an existing deck with thumbnails and markitdown, map each content section to an appropriate layout, unpack the package, reshape the slide list, then fill placeholders in slide XML. The guidance stresses visual variety—multi-column, image-led, quote, and stat layouts—because monotonous bullets are called out as a common failure mode. You run local Python helpers for unpacking and duplication, complete structural changes before touching copy, and can parallelize content edits across slides when your agent supports subtasks. It fits when you already have a .pptx template or prior deck to adapt for pitches, product walkthroughs, or stakeholder updates, not when you only want a one-shot narrative with no file to mutate.607installs16DocxThe docx skill in k-dense-ai’s scientific-agent-skills collection equips coding agents to handle Microsoft Word documents as first-class artifacts. Solo builders and small research teams often need polished .docx outputs—grants, lab reports, SOPs, or client memos—without leaving the agent loop. The SKILL.md positions the capability around professional document editing patterns, including operations familiar from Word’s review workflow such as managing tracked changes. Because the pack is labeled for scientific agents, it pairs naturally with literature workflows, citation skills, and manuscript preparation elsewhere in the repo. Install it when your pipeline must produce or iterate on binary Office files rather than markdown-only drafts. Expect intermediate familiarity with document structure, revision markup, and whatever local libraries or tooling the skill invokes for OOXML.605installs17Exploratory Data AnalysisExploratory Data Analysis is a scientific-agent skill that guides your coding assistant through a consistent, publication-style EDA report for any dataset file you drop into the repo. Solo builders use it when a CSV, HDF5, or specialty format arrives without documentation and you need file identification, library recommendations, missing-value accounting, and sanity checks before training or shipping analytics features. The markdown template forces executive summary, structural stats, validity gates, and optional temporal properties so outputs are citable in plans and PRs. It fits Validate prototype work first, then carries into Build backend or data-pipeline tasks when you harden schemas. Complexity is intermediate because you must interpret domain ranges yourself; the skill structures the investigation rather than auto-running every library.603installs18SeabornSeaborn is a phase-specific agent skill that packages common statistical visualization recipes for solo builders and researchers working in Python. It focuses on practical snippets you can adapt during exploratory data analysis—pairwise relationship plots, distribution exploration across conditions and timepoints, and annotated correlation heatmaps with upper-triangle masks—as well as patterns aimed at scientific publications such as shared axes, publication styling hooks, and saving figures at print-friendly resolution. The skill assumes familiarity with pandas DataFrames and matplotlib’s pyplot interface; it does not replace Seaborn’s full API reference but accelerates agent-generated code when you need correct parameter combinations for displot, pairplot, heatmap, and faceting without re-reading docs each time. Reach for it while building analytics dashboards, research notebooks, internal metrics tools, or ML experiment reports where clear, statistically honest charts matter.602installs19Database LookupDatabase-lookup packages procedural knowledge for querying public scientific databases from an AI coding agent—starting with Addgene for plasmid metadata and AlphaFold DB for predicted protein structures. Solo builders in computational biology, biotech side projects, or research tooling can avoid re-deriving base URLs, auth headers, and file naming conventions every time they spawn a new chat session. The skill spells out Token authorization for Addgene, example GET paths for search and detail endpoints, and the predictable AlphaFold file URL pattern for PDB, mmCIF, and predicted aligned error JSON. Prism shelves it under Idea → research because the immediate value is discovery and citation during opportunity and scope work, though the same calls often reappear when you wire Build → integrations or Grow → content that cites structures. It is not a full LIMS or warehouse; it is a lookup integration layer. Pair it with your own caching or compliance review before production use of third-party sequence data.601installs20Bgpt Paper SearchBGPT Paper Search is an agent skill that wires your coding agent to the BGPT remote MCP server so you can query a curated corpus of papers with data extracted from full text, not just metadata. Solo builders and indie researchers use it when they need reproducible evidence for product claims, technical choices, or regulatory narratives without manually opening dozens of PDFs. The skill fits early journey work—scoping a health or science-adjacent feature, validating whether an approach is supported by trials, or drafting docs that cite methods and sample sizes accurately. Invocation assumes the MCP server is configured in the host, network access to bgpt.pro, and optionally a BGPT API key. Compared with generic web search, the emphasis is on structured experimental payloads suitable for synthesis tables and agent-side reasoning, which makes it especially useful when your agent must compare studies on equal footing.595installs21StatsmodelsStatsmodels is an agent skill that packages a discrete choice modeling reference for solo builders who need rigorous inference without hiring a statistician full time. It explains when to apply binary logit, multinomial, ordinal, and count formulations, walks through standard statsmodels APIs such as Logit with constant terms, and emphasizes interpretation via odds ratios, confidence intervals, and marginal effects rather than raw coefficients alone. The canonical journey placement is Validate during scoping: you are proving whether signup drivers, pricing tiers, or feature adoption categories behave as your narrative claims. The same patterns recur in Grow when analyzing lifecycle segments and in Build when embedding calibrated predictors into a backend service. Treat it as intermediate-to-advanced procedural knowledge grounded in maximum likelihood and explicit distributional assumptions, so your agent produces reproducible Python snippets and readable statistical summaries aligned to business questions.592installs22Hypothesis GenerationHypothesis Generation is an agent skill for solo builders and indie researchers who need publication-quality hypothesis reports without hand-rolling LaTeX every time. It packages a dedicated hypothesis_generation style file, custom summary and evidence boxes, and a consistent color system so multiple competing hypotheses read clearly on the page. You use it when you have defined a phenomenon and want numbered hypotheses, testable predictions, supporting evidence, and limitations in one shareable PDF. The skill targets Markdown-first agent workflows that still need a formal deliverable for advisors, collaborators, or your own decision log. Compilation assumes XeLaTeX or LuaLaTeX for font and box fidelity. It complements literature ingestion skills by focusing on structuring what you believe and how you would falsify it—not on collecting raw papers.591installs23Scholar Evaluationscholar-evaluation embeds the ScholarEval framework so your coding agent can judge scholarly material with explicit rubrics instead of vague “looks fine” feedback. Solo builders and indie researchers use it when turning messy notes into a defensible research question, stress-testing an agent-written lit review, or calibrating a paper or grant section before validation experiments. Each dimension—including problem formulation and research questions—uses five anchored levels from Excellent to Poor with concrete indicators for specificity, gap significance, scope feasibility, differentiated contribution, and justified impact. The skill fits multi-phase journeys: sharpen questions during Idea research, tighten scope during Validate, and re-run checks on Build-phase docs or agent reports that cite literature. It does not fetch papers or run statistics; it supplies evaluation criteria and quality language so human and agent reviewers align on what “good” means.591installs24MarkitdownMarkItDown is an agent skill that teaches your coding agent to use Microsoft’s MarkItDown Python library to turn binary office and PDF files into clean Markdown text. Solo builders use it when literature lives in PDFs and slide decks but prompts and RAG pipelines need plain text. The skill documents one-off conversion, saving to .md files, stream-based reads, and batch conversion across a papers directory—patterns that fit indie research, competitive intel, and internal doc ingestion without a separate ETL service. It pairs naturally with hypothesis generation, note-taking, and agent tooling that consumes Markdown corpora. Complexity stays beginner-friendly: import the library, call convert, write output. You still own licensing and copyright for sources you convert. For production agents, filesystem access is the main requirement; no browser or network is mandated by the examples themselves.590installs25Latex PostersLatex-posters is an agent skill that ships a professional baposter-based research poster skeleton for solo builders and indie researchers who need print-scale layouts without wrestling LaTeX poster classes alone. It targets the build phase documentation track when you turn analysis, benchmarks, or product research into a single A0 PDF for lab meetings, meetups, or academic venues. The template encodes column count, spacing, colors, and common scientific packages so your coding agent can drop in figures, methods, and QR links while preserving poster-specific typography scaling. Use it when you already have narrative and assets but lack a repeatable poster scaffold. It pairs well with other scientific-writing skills in the same repo and keeps output vendor-neutral compared to Canva-only workflows, while still expecting you to compile locally or in CI.583installs26Pptx Posterspptx-posters is a specialized agent skill for research posters built with HTML and CSS, exportable to PDF or converted to PowerPoint when editors need PPT workflows. Solo builders and researchers should invoke it only after they explicitly ask for a PPTX poster, PowerPoint poster, HTML-based poster, or a non-LaTeX path when LaTeX is unavailable. For generic poster requests, the skill itself directs you to latex-posters for better typographic control at academic conferences. The web-based approach favors fast iteration, responsive sections, and straightforward visual placement before print or PPTX conversion. It sits in the launch phase as a distribution artifact for talks and poster sessions, not as everyday product documentation. Beginners can follow the export steps; conference typography purists may still prefer LaTeX unless PowerPoint editing is a hard requirement.583installs27Research Grantsresearch-grants is an agent skill from k-dense-ai’s scientific-agent-skills set that walks you through writing a budget justification—the narrative reviewers expect beside the spreadsheet. It emphasizes that each dollar must map to proposed work: senior personnel effort in calendar months, inflation and escalation across award years, and plain-language duty statements tied to specific aims. Solo academic founders, indie lab leads, and small-team PIs use it when a sponsor’s template is vague and you need consistent sections without reinventing NIH-style phrasing each cycle. The skill is template-driven documentation, not grant discovery or compliance filing; pair it with your internal budget model and agency guidance. Multi-phase placement reflects scoping during Validate, drafting during Build docs, and final polish before Ship-style submission packages.576installs28Generate Imagegenerate-image is an agent skill that teaches coding agents to produce and edit raster images through OpenRouter’s image-capable models using a small Python entry script. Solo builders use it when they need hero art, diagrams, dataset illustrations, or marketing visuals without leaving the terminal or agent session. The flow covers API key discovery in .env files, model selection across providers like Google Gemini image preview and Black Forest Labs Flux variants, and optional edit mode when an input image accompanies the prompt. It fits scientific and indie workflows where reproducible CLI calls matter more than a GUI. You should have Python available, an OpenRouter account, and comfort storing secrets locally. Outputs are image files or encoded payloads from successful API runs—not a hosted gallery or design system. Pair it with content or launch tasks when assets are ready; during Build it accelerates placeholder and final media generation for agents and small SaaS products.567installs29Pytorch LightningPyTorch Lightning best practices is an agent skill that teaches solo builders and small teams how to organize trainable models the way the Lightning framework expects: research code in LightningModule hooks, reproducible data in LightningDataModule, and scale-out settings in Trainer. It matters when ad-hoc training scripts quietly reimplement device placement, DDP, and dataloaders—and become brittle the first time you add a second GPU or a validation split. The skill walks through what to do (compute loss in training_step, return it; let Trainer own the loop) and what to avoid (manual .cuda(), optimizer steps in the step unless you opted into manual optimization). It also steers you toward prepare_data for one-time downloads and setup for per-rank dataset construction, which is the usual footgun in distributed jobs. Install it when you are building ML backends, fine-tuning models for agent features, or shipping experiment code you want an coding agent to extend without breaking distributed training.565installs30InfographicsInfographics is a reference skill from scientific agent skills that catalogs colorblind-safe and professional palette options for infographic and data-visualization work. Solo builders and researchers use it when asking an agent to illustrate papers, dashboards, landing explainers, or internal docs without picking colors that fail accessibility or print poorly. The readme centers on Wong's widely cited seven-color set and IBM's eight-color accessible palette, each with hex, RGB, and suggested usage roles such as primary data, alert, and highlight. Prompt-ready phrasing lets you constrain image or diagram generation to specific hex values. It does not run tools itself; it supplies authoritative palette choices so downstream design, docs, and content workflows stay consistent. Best paired with generative media or slide-building tasks where you need defensible color science rather than aesthetic guesswork.562installs31NetworkxThe networkx skill is procedural documentation for NetworkX graph algorithms that solo builders drop into Python backends, data pipelines, or agent tools. It covers shortest-path families—from Dijkstra and Bellman-Ford for weighted graphs through A* with custom heuristics—to connectivity analysis such as component counts, largest connected subgraphs, and node-local membership. All-pairs helpers and average shortest path length support analytics dashboards or research scripts without re-deriving API names. The readme assumes you already hold a NetworkX graph object and need correct function choice for negative weights, heuristics, or undirected connectivity. Intermediate complexity: graph theory basics help, but the skill is primarily a curated snippet catalog for coding agents implementing routing, dependency graphs, or network metrics in a indie SaaS or internal CLI.560installs32TransformersThe transformers skill teaches solo builders how to run text generation with Hugging Face causal language models using `model.generate()` after tokenizing inputs to PyTorch tensors. It walks through loading GPT-2-style checkpoints, decoding outputs with `skip_special_tokens`, and choosing among greedy decoding, stochastic sampling, and beam search depending on whether you need determinism or variety. The readme explicitly positions the Pipeline API as a fast wrapper while reserving direct generation for custom preprocessing and decoding control—typical when you ship an agent feature, API endpoint, or batch inference job rather than a one-off notebook. Expect intermediate familiarity with Python, tensors, and generation hyperparameters. The content is procedural reference material your coding agent can follow while implementing LLM features in a solo SaaS or internal tool.560installs33Optimize For Gpuoptimize-for-gpu is an agent skill for solo and indie builders who ship scientific or data-heavy Python and need real speed without becoming CUDA experts. It applies when you mention GPU, CUDA, NVIDIA, or when large NumPy, pandas, scikit-learn, NetworkX, GeoPandas, Faiss, or scipy.sparse workloads sit on the critical path. The skill walks a six-step workflow: profile on CPU, match the problem to the right NVIDIA library (array math via CuPy, kernels via Numba or Warp, DataFrames via cuDF, ML via cuML, graphs via cuGraph, imaging via cuCIM, search via cuVS, storage via KvikIO, and more), refactor hot loops and memory transfers, benchmark against baseline, and re-check numerical correctness. It is written for builders using Claude Code, Cursor, or Codex in a repo—not a hosted GPU service—and pairs well with profiling before you rewrite entire pipelines.559installs34AeonAeon is an agent skill documenting the Aeon library’s anomaly-detection toolkit for scientific and product time series. Solo builders running metrics, experiments, or sensor feeds can use it to pick the right algorithm class instead of defaulting to a single Isolation Forest call. The skill distinguishes collection-level detection—finding which entire series look wrong—from within-series detection that locates discordant subsequences or point outliers using matrix-profile and clustering approaches. Each listed estimator includes practical routing hints, such as training classifiers on normal data, using K-means distance for well-clustered normals, or choosing incremental STAMP variants for streaming feeds. Install it when you are wiring observability, batch analytics jobs, or research notebooks where false positives and method fit matter. It assumes Python scientific stack familiarity and complements generic monitoring by focusing on statistical time-series anomaly patterns.558installs35Get Available ResourcesGet Available Resources is an agent skill built around a Python detection script that snapshots the machine your coding agent is running on. Solo builders doing scientific computing, bioinformatics, or large tabular workflows install it so the agent does not guess whether parallel frameworks or out-of-core formats are appropriate. The script collects physical and logical CPU counts, memory and swap totals, disk availability, and GPU backends across NVIDIA, AMD, and Apple Silicon, then writes structured JSON the agent can cite in plans. It explicitly nudges decisions between Dask, Zarr, Joblib, and related approaches based on headroom rather than defaults. Use it at the start of a heavy notebook refactor, a batch pipeline task, or when moving a prototype from a laptop to a workstation. Prerequisites are Python with psutil and permission to run local shell commands in the agent environment.557installs36PaperzillaPaperzilla is an agent skill that connects your assistant to Paperzilla projects, recommendations, and canonical papers using the official `pz` CLI (Homebrew on macOS, Scoop on Windows, or Linux install docs). Solo and indie builders running scientific or AI-heavy work install it when they want conversational access to recent project recommendations, deep dives on why a recommendation matters, markdown-based paper fetch and summary, and feed export—without baking a rigid multi-step ritual into the repo. The skill explicitly does not impose a workflow; it is direct data access your agent can combine with brainstorming, specs, or implementation elsewhere. Typical triggers match user asks for latest recommendations, canonical paper details, feed URLs, and feedback on recommendations. You need network access, a working `pz` install, and whatever auth your Paperzilla profile requires per their CLI getting-started guide.556installs37Market Research ReportsMarket Research Reports is a documentation-oriented agent skill that teaches how to format investor- and client-grade market studies with the market_research.sty LaTeX package. Solo founders validating a niche, indie consultants, and small teams shipping PDF deliverables use it when raw research must become a consistent visual system: branded blues for titles, green market-data panels, orange risk warnings, and purple recommendation boxes. The SKILL.md is a formatting guide—not a data vendor—so your agent applies the right environments and color tokens while you supply the actual market statistics and citations. It pairs naturally with research skills that gather TAM, competitor, and segment facts, then outputs structured LaTeX sections ready for compile. Intermediate complexity assumes basic LaTeX toolchain comfort. Skip if you only need Notion or Google Docs memos without stylized PDFs. Confirm package provenance via Prism Security Audits before piping generated LaTeX through automated compile jobs on CI.554installs38MatlabMATLAB (Data Import and Export Reference) is a build-phase reference skill for solo builders and small teams who still rely on MATLAB for lab, engineering, or legacy analytics pipelines. It gives copy-ready patterns for ingesting mixed-type CSVs as tables or matrices, tuning import options, and exporting results without guessing delimiter or header behavior. Coverage spans high-level table APIs, cell and string line reads, MAT persistence, image I/O, and lower-level file operations—enough to unblock an agent implementing a reproducible script rather than improvising syntax. Use it when you are wiring datasets into models, merging spreadsheet inputs, or standardizing exports for downstream Python or BI tools. It does not replace domain modeling or statistical design; it keeps file boundaries correct so your computation layer stays trustworthy.554installs39ShapSHAP Explainers Reference is an agent skill that packages procedural knowledge for the SHAP library: which explainer to instantiate, what masker and background data mean, and how algorithms differ across tree models, black-box models, and Python callables. Solo and indie builders shipping ML-backed SaaS, internal APIs, or scientific agents use it when a stakeholder asks why a score changed, when debugging fairness or drift, or when wiring explainability into a batch scoring job without rereading upstream docs. The skill emphasizes the general shap.Explainer auto-selector as the default, then dives into TreeExplainer for ensemble speed and exactness, including feature_perturbation options such as interventional versus tree_path_dependent behavior. It is reference depth, not a one-click audit: you still need labeled data, a trained model, and judgment about computational cost. Intermediate complexity fits builders who already run scikit-learn or gradient-boosted pipelines and want their coding agent to generate correct constructor calls and method usage the first time.554installs40Polarspolars is a scientific-agent reference skill that encodes Polars best practices so solo builders ship faster, correct dataframe pipelines without relearning optimization folklore. It explains when to prefer lazy frames, how to push filters and column selection upstream, and why staying in the expression API matters for parallel execution on larger files. The material is framed as do-this-not-that examples rather than a one-off script, which makes it useful whenever an agent is implementing analytics backends, research notebooks turned into jobs, or ETL steps in a SaaS product. Prism catalogs it for builders who want procedural knowledge in SKILL.md form—invocable during Build, but equally relevant when you Operate on slow queries or Grow analytics features.553installs41DaskDask Array is an agent skill that teaches solo builders and small teams how to use blocked NumPy-compatible arrays for datasets that do not fit in memory. It explains the chunk grid model, which mathematical and reduction operations are supported, and when parallel out-of-core work beats staying on vanilla NumPy. The skill fits indie hackers shipping analytics features, scientific tooling, or batch jobs in Python who already know NumPy but hit RAM limits. Use it during backend and pipeline implementation when profiling shows memory pressure or when you need multi-core parallelism without rewriting your array idioms. It does not replace a full cluster operations playbook; it focuses on the array API and scaling patterns so your agent can suggest Dask structure, chunk sizing intuition, and interoperability with the NumPy ecosystem.552installs42PyzoteroPyzotero is a scientific-agent integration skill that teaches solo builders and researchers how to wire coding agents to Zotero without leaking credentials. It documents where to obtain a user ID and API key on zotero.org, how to target group libraries from URL paths, and the recommended environment-variable layout loaded through python-dotenv before constructing a pyzotero.Zotero client. Separate snippets distinguish personal user libraries from group libraries, using the group numeric ID as library_id with library_type appropriate to the collection. The security section is explicit: placeholders in docs are illustrative, production keys belong only in environment variables or a narrowly scoped .env file, never in committed source. For indie builders validating ideas, writing docs with citations, or automating literature review inside Claude Code or Cursor, this skill is the setup layer before fetch/search automation. It assumes Python and network access to Zotero’s API; it does not replace reference-manager UX in the browser.551installs43Open NotebookOpen-notebook is an agent skill that exposes the Open Notebook HTTP API so coding agents can manage research notebooks programmatically instead of clicking through a UI. Solo builders and scientist-indies use it when a local Open Notebook instance (default port 5055) backs their literature review, experiment log, or RAG source library. The reference covers authentication when a password is set, which routes stay public, and CRUD operations on notebooks including archival filters, ordering, and safe delete previews. It pairs with scientific-agent workflows where sources and notes must stay structured for later synthesis or agent retrieval. Complexity is intermediate because you must run the server, understand REST verbs, and respect auth headers. It is an integration skill pattern: the deliverable is correct API calls and consistent notebook state, not generated prose summaries alone.550installs44Parallel WebParallel Web (documented in-repo as data enrichment) is an agent skill that walks your coding agent through starting Parallel enrichment runs: inline JSON batches or CSV uploads, a natural-language intent describing fields to add, and asynchronous execution so the session does not hang. Solo builders use it when competitor lists, lead sheets, or research tables have gaps—founding year, leadership, firmographics—and manual tab research does not scale. The skill emphasizes operational details easy to get wrong: always pass --no-wait, write targets to a known path, and when continuing a prior Parallel research or enrich turn, chain interaction_id so context carries forward without re-explaining earlier findings. It is intermediate complexity because you must substitute real column names and paths, but the command patterns are copy-pasteable. Prism places it on Idea/research first while acknowledging reuse in Grow when you refresh lifecycle or analytics tables from the same CLI.550installs45Scikit Survivalscikit-survival is an agent skill that teaches competing-risks survival analysis for solo builders and small teams running cohort studies, clinical-style outcomes, or operational time-to-event data. It explains when multiple mutually exclusive event types make standard Kaplan–Meier misleading and walks through the Cumulative Incidence Function as the correct probability of a specific event by time t. The skill maps concrete domains—transplant infections, job termination causes, equipment failure modes—and states clear guardrails: use competing risks when events block each other, skip when only one event matters or events recur. It is aimed at builders wiring Python or scientific-agent workflows who need statistically defensible incidence estimates and covariate interpretation without treating competing events as independent censoring.549installs46SympySymPy Advanced Topics is an agent skill that documents how to use SymPy for geometry, number theory, combinatorics, logic, statistics, polynomials, and special functions in Python. Solo builders shipping research tools, education apps, or scientific SaaS can point Claude Code or Cursor at these patterns so the agent writes correct symbolic code instead of guessing APIs. The skill is reference-oriented: copy-pasteable blocks for 2D shapes, segment midpoints, circle metrics, and intersection queries, plus pointers into deeper SymPy domains. Use it during implementation when a feature needs exact math, symbolic simplification, or reproducible geometric reasoning. It does not replace a full computer-algebra course, but it shortens the loop from spec to working sympy expressions for indie-sized codebases.545installs47Pymcpymc is a scientific agent skill that ships a structured template for Bayesian hierarchical and multilevel models using PyMC and ArviZ. Solo builders and indie researchers use it when observational data has natural groups—cohorts, sites, or SKUs—and a flat regression would hide partial pooling benefits. The readme walks through data preparation, prior choices, model building, inference, and visualization, with explicit TODO anchors so the agent replaces synthetic demo data with real CSVs. It assumes comfort with probabilistic programming and Python scientific stack packages, and it prioritizes correct grouped indexing over production serving. Reach for it during validation when you need credible uncertainty on group-level effects before committing product or policy decisions.543installs48Torch GeometricTorch Geometric is a scientific agent skill that documents how to load graph machine-learning data with PyTorch Geometric—either ad hoc lists of `Data` nodes and edges or reusable `InMemoryDataset` subclasses with download and process pipelines. Solo builders doing recommendation, molecule, or knowledge-graph experiments use it when raw inputs live in CSV, pandas, or numpy and must become `x`, `edge_index`, and `y` tensors batched for training. The reference covers skipping heavy dataset machinery for synthetic runs, implementing the four lifecycle methods for RAM-sized corpora, and using trusted downloads with checksum discipline. In Prism it sits in the ML build lane for agents extending the k-dense scientific skill pack, complementing model code rather replacing experiment design or hyperparameter search.543installs49HypogenicHypoGeniC is a configuration template for scientific agent workflows that generate and refine testable hypotheses from labeled text datasets. Solo builders and small research teams use it when they need repeatable experiment configs—data paths, model settings, cache, and generation method switches—rather than one-off chat prompts. The template documents train/validation/test JSON expectations (text feature lists plus labels), supports hypogenic, hyporefine, and union flows, and wires optional Redis to cut API spend during iterative runs. It fits the idea and validate phases when you are exploring classifiers or annotation schemes before locking an implementation plan. It is not a runnable skill by itself; you pair it with the broader scientific-agent stack and your own data. Expect intermediate familiarity with ML splits, env-based API keys, and batch iteration limits.542installs50Scikit Bioscikit-bio is an agent skill that packages a detailed API reference for the scikit-bio Python library used in computational biology and microbiome research. Solo builders and small teams shipping analysis tools, research agents, or data products install it when they need correct imports, sequence operations, alignment calls, and diversity workflows without hallucinating method names. The skill mirrors official-style usage for DNA/RNA/protein objects, tree and ordination utilities, and common I/O paths. It fits the Build phase when you integrate scientific Python into scripts, CLIs, or backend jobs rather than during pure market validation. It is reference-heavy intermediate material: you should already know your biological question and test strategy. Prism lists it so agents working on niche science stacks have procedural knowledge comparable to skills.sh task skills, optimized for lookup during implementation.541installs51Umap LearnUMAP Learn is a scientific agent skill that compresses the umap-learn 0.5.12 Python API into agent-friendly reference material for manifold learning and 2D/3D embeddings. Solo builders and indie data hackers use it when exploring high-dimensional user, log, or feature data and need correct parameter choices—neighborhood size, distance metrics, initialization, and memory behavior—without rereading the full upstream docs. It supports workflows from exploratory notebooks through backend scripts that feed dashboards or clustering steps. The content emphasizes practical tuning bands and constructor defaults so agents do not hallucinate removed or renamed arguments. Pair it with your eval pipeline when comparing embeddings for search, segmentation, or QC visuals.541installs52BiopythonBiopython is a scientific agent skill focused on advanced Biopython features for solo builders and researchers shipping computational biology tooling. It walks through creating sequence motifs from instance sets, computing consensus and degenerate consensus, building normalized PWMs with pseudocounts, deriving information content, and searching sequences with PSSM log-odds thresholds. It also covers reading motifs from JASPAR-format files and iterating parsed motif collections. Use it when your agent needs correct, library-native patterns instead of reinventing motif math or file parsers. The skill is phase-specific to building analysis backends, CLIs, or APIs that consume biological sequence data—typical for indie bioinformatics utilities, research prototypes, and data pipelines rather than marketing or deploy automation.539installs53Timesfm ForecastingTimesFM Forecasting is an agent skill backed by a runnable Python example that combines classical context anomaly detection with Google TimesFM quantile forecast intervals. Solo builders use it when they need defensible time-series monitoring—linear detrend and Z-scores on a historical window, then forward-looking bands on a horizon (default 12 steps)—without hand-rolling every statistical step in chat. The sample script loads real NOAA global temperature anomaly data, flags a documented critical point, injects anomalies into a synthetic future segment, and emits both a two-panel chart and JSON detection records. It suits agent projects that must ship reproducible notebooks or scripts, SaaS ops dashboards fed by batch jobs, and API services that surface forecast envelopes. Expect intermediate comfort with Python, NumPy, Pandas, and Matplotlib plus GPU/CPU environment setup for TimesFM itself.538installs54Consciousness CouncilConsciousness Council is a journey-wide agent skill reference for configuring specialized multi-member councils that deliberate on hard decisions from multiple staged perspectives. Rather than one assistant voice, you assemble rosters tuned to the domain—startup bets with Strategist and Contrarian tension, architecture reviews pitting Architect against Minimalist, hiring choices weighing Empath and Ethicist, or creative direction with Creator and Outsider. Each configuration documents why those personas matter and which disagreement to surface so the final synthesis is tested, not polite. Solo builders use it whenever a choice has irreversible cost: market positioning, system shape, key hire, or brand direction. It complements generic agent skills by supplying advanced council recipes; you still supply the actual question and constraints. The skill does not fetch data or run votes automatically—it structures how your agent should role-play and reconcile opposing views before you commit.537installs55Venue TemplatesVenue-templates is a procedural template skill for scientific agents that need Cell Press–specific front matter: Summary (abstract), Highlights, and electronic table-of-contents blurbs. Solo and indie builders running research or bioinformatics agents use it when a draft needs venue-accurate length limits and rhetorical shape without re-reading author guidelines. The bundled examples walk through a full senescence paper snippet so the model mirrors real publisher constraints—word caps, bullet character limits, and outcome-forward closing sentences. It does not run experiments or cite literature; it standardizes how findings are compressed for editors and readers. Pair it with broader scientific-writing or figure skills when you already have results and need submission-ready packaging for Cell-family journals.536installs56What If OracleWhat-If Oracle is a reference skill of domain-specific scenario analysis templates for agent-assisted decision work. It does not run a single API; it tells your coding agent how to frame what-if questions across startup moves, architecture bets, financial exposure, and career pivots, including which analytic branches to weight and copy-paste prompt scaffolds. Solo and indie builders use it when a choice has second-order effects and they need consistent structure instead of one-off brainstorming. Pair it with broader scientific or brainstorming skills in the same repo when you want quantified branches; use these templates as the checklist for variables, emphasis, and narrative constraints before you lock a plan or implementation path.536installs57Stable Baselines3stable-baselines3 is a reference skill that helps solo builders and ML tinkerers choose among Stable Baselines3 reinforcement-learning algorithms before they invest days in the wrong trainer. It centers on a comparison matrix linking algorithm type (on- versus off-policy), supported action spaces, sample efficiency, training speed, and typical use cases—from PPO as a stable generalist through SAC and TD3 for continuous control, DQN for discrete domains, HER for goal-conditioned sparse rewards, and RecurrentPPO for partial observability. Deeper sections summarize strengths such as PPO’s vectorized-environment fit and SAC’s sample efficiency on continuous tasks. It is aimed at agent builders prototyping robotics, game AI, scheduling bots, or simulation policies in Python, not at production MLOps deployment alone. Invoke it when you have an environment spec (discrete vs continuous actions, observability, goal structure) and need a justified starting algorithm and training expectations rather than trial-and-error across every SB3 class.535installs58GeopandasGeopandas is a scientific agent skill that gives agents procedural knowledge for coordinate reference systems in GeoPandas. It walks through understanding CRS as pyproj.CRS objects, checking whether CRS is defined on a GeoDataFrame, and the critical distinction between set_crs—which only attaches metadata when coordinates are already correct—and to_crs—which actually reprojects geometry. The readme enumerates accepted CRS formats via pyproj.CRS.from_user_input(), including EPSG codes, ESRI authorities, WKT, and PROJ strings, plus practical reprojection examples such as EPSG:4326 to EPSG:3857 Web Mercator. Solo builders shipping location-aware analytics, climate tooling, logistics dashboards, or research notebooks install it so agents do not silently mislabel or warp geometries. It is reference-grade guidance rather than a one-click ETL job, meant to be applied whenever spatial joins, area calculations, or map overlays depend on consistent CRS.531installs59BioservicesBioServices is an agent skill that teaches identifier mapping across major biological databases using the BioServices Python library. Indie builders and small bioinformatics teams use it when pipelines, research agents, or internal tools must translate UniProt accession numbers into KEGG gene IDs, link compounds via UniChem, or batch-normalize IDs before analysis. The skill is reference-heavy: it walks through UniProt mapping parameters, batch queries, KEGG conventions, and troubleshooting when cross-database lookups fail. It sits in Build because the outcome is integration code you run repeatedly—not a one-time literature review. Pair it with notebooks or ETL jobs where consistent identifiers prevent silent join errors. The guide assumes you are comfortable calling external biology APIs and handling heterogeneous ID namespaces in scientific software.529installs60PymooPymoo is an agent skill that acts as a compact algorithms reference for the pymoo optimization library, aimed at builders running scientific, ML, or engineering workloads who need the right evolutionary or swarm method without rereading full documentation. It catalogs single-objective options including Genetic Algorithm, Differential Evolution, Particle Swarm Optimization, and CMA-ES, with purpose statements, best-for problem shapes, and concrete parameters like population size, selection, crossover, and mutation defaults for GA. The skill supplies copy-paste instantiation patterns so an agent can propose algorithm = GA(pop_size=100, eliminate_duplicates=True) and explain when DE or PSO is a better global search bet on continuous spaces. Use it during backend implementation when you are encoding decision variables, constraints, and fitness evaluation and must choose a solver family before benchmarking. It is reference material for algorithm choice and parameter knobs, not a hosted optimizer service. Pair it with your own problem definition code and pymoo installed in the project environment.529installs61Ggetgget is an agent skill that equips solo builders in computational biology with accurate context for the gget Python package’s database integrations. Rather than guessing which Ensembl release or UniProt endpoint applies, the skill summarizes what each module queries, how often sources update, and reproducibility practices like specifying Ensembl release numbers or common species shortcuts. It fits builders creating CLIs, pipelines, or research agents that fetch references, search annotations, or run BLAT alignments. Advanced complexity reflects shifting external schemas and domain-specific nomenclature. Use during implementation when your agent writes or debugs gget calls; skip if you are not working with genomics or protein data. Prism classifies it as phase-specific build integrations because it anchors on live scientific APIs, not journey-wide methodology.528installs62ModalModal is an agent skill that equips solo builders and small teams to deploy and call serverless Python on Modal’s autoscaling container pools. It maps the core object model—App for grouping, Function for stateless workers, Cls for lifecycle-aware classes—and the execution surface you use day to day: synchronous remote runs, async spawn, and parallel map-style workloads. The reference also highlights operational knobs that matter when costs and latency bite: GPU tiers, input concurrency, batching windows, and live autoscaler tweaks. Install it when you are moving notebooks or scripts into durable cloud endpoints, wiring scientific or agent workloads that need burst GPU, or debugging how Modal’s decorators and invocation helpers differ from ordinary local Python. It is reference material, not a deploy script: pair it with your own infra choices and security review before production traffic.528installs63Simpysimpy is an agent skill that teaches SimPy’s event system for discrete-event simulation in Python. Solo builders and small teams use it when they need to model queues, timed processes, or resource contention without building a custom scheduler. The guide walks through event states, triggering with succeed(), and the most common pattern—yielding env.timeout to advance simulated clock time and optionally pass results back into processes. It is aimed at developers who already write generator-based processes and want correct mental models before layering stores, resources, or monitoring. Invoke it while designing or debugging simulation logic so agents generate idiomatic SimPy rather than imperative loops with manual timers.528installs64VaexVaex is an agent skill that teaches agents how to work with Vaex DataFrames for large tabular scientific and analytics workloads. Solo builders shipping data pipelines, ML feature stores, or internal research tools install it when pandas runs out of memory or when opening multi-gigabyte files must stay instant. The skill covers DataFrame fundamentals, lazy evaluation, opening HDF5 and Arrow (recommended), Parquet, lazy CSV since 4.14, FITS, and wildcard patterns, plus format-specific loaders. It contrasts Vaex with pandas on memory and execution model so agents choose the right API for exploration versus production ETL. Best used during backend and analytics implementation when datasets exceed comfortable RAM limits.527installs65Clinical ReportsClinical Reports is a structured writing template skill for agents helping clinicians and medical researchers produce case reports that match common journal expectations. It walks through title conventions, author blocks, MeSH-oriented keywords, and a multi-part abstract (introduction, patient concerns, diagnosis, interventions, outcomes, lessons) before expanding into background and patient-information sections. Solo builders here are often indie researchers, residents, or clinician-scientists who use AI assistants to accelerate first drafts while keeping human oversight for PHI, ethics, and institutional review. The skill does not diagnose patients or replace IRB processes; it standardizes prose so you spend editing time on accuracy and consent rather than reinventing section headings. Use it when you already have de-identified clinical facts and need a consistent manuscript outline the agent can populate iteratively.526installs66AnndataAnnData is an agent skill that encodes library-specific best practices for annotated matrix containers used heavily in single-cell genomics and other high-dimensional experiments. Solo builders and small teams shipping Python analysis repos install it when agents otherwise default to dense numpy arrays, string-heavy metadata columns, or full in-memory loads that blow RAM on real H5AD files. The guidance walks through sparsity checks, categorical encoding for obs and var, backed read modes for large on-disk datasets, and the view/copy semantics that cause subtle bugs during subsetting. It is aimed at developers who already use NumPy, SciPy sparse, and anndata rather than newcomers learning biology from scratch. Use it during implementation of preprocessing, QC filtering, and pipeline refactors where performance and correctness of AnnData objects matter as much as the biology.525installs67Clinical Decision SupportClinical Decision Support is an agent skill from a scientific skills collection oriented toward producing genomic profile style reports suitable for clinical decision-support narratives. The bundled material centers on a formal LaTeX document template: branded headers, tier-colored boxes, and distinct visual treatment for mutation, amplification, and fusion findings. Solo builders working on health-adjacent or bioinformatics agents can use it when they need consistent, citable report structure rather than free-form chat summaries. It sits in Build because the primary job is authoring a regulated-looking artifact from structured inputs, not running trials or operating production monitors. Advanced users should treat outputs as drafts requiring qualified clinical review and local compliance rules. Pair it with your own data pipelines and validation processes; the skill supplies document pattern and section discipline inferred from the template, not verified medical advice.524installs68GeomasterGeoMaster is a deep geospatial science skill for solo builders and small teams building location-aware analytics, Earth-observation pipelines, or research prototypes. It organizes remote sensing, vector GIS, geostatistics, and machine learning for imagery into installable quick starts and eleven reference documents—from core Python stacks through JavaScript and Julia tooling. You can trace satellite mission context to preprocessing steps, pick QGIS versus scriptable GDAL flows, and follow big-data and GPU acceleration notes when desktop GeoPandas stops scaling. The skill suits indie climate tools, agritech dashboards, civic mapping side projects, and agent-assisted notebooks where the agent must not hallucinate library APIs. Expect advanced breadth rather than a single happy-path tutorial; pair it with your dataset licenses and compute budget before production ship.524installs69TorchdrugTorchDrug is a PyTorch-oriented toolkit for graph-based machine learning in drug discovery and related scientific domains. This agent skill distills core architecture—how representation models, task objects, datasets, and the Configurable base class fit together—so solo builders and small teams can stand up experiments faster inside Claude Code, Cursor, or Codex. It is aimed at builders who already work in Python ML and need procedural knowledge rather than a one-off script: when to separate encoders from objectives, how to persist full pipelines as configs, and how to swap components without rewriting glue code. Use it while designing backends, agent tooling for research automation, or validation prototypes that depend on molecular or knowledge graphs. It does not replace hands-on benchmarking, GPU ops, or regulatory validation of any therapeutic claims.522installs70PyhealthPyHealth is an agent skill for building clinical and healthcare deep-learning workflows with the PyHealth toolkit. Solo builders and indie teams install it when they need a consistent path from raw or benchmark clinical data to trained models and reportable metrics, even if the user says MIMIC, eICU, OMOP, or drug recommendation without naming PyHealth explicitly. The skill encodes the library’s stable interfaces: pick a dataset loader, bind a clinical prediction or coding task, instantiate an appropriate architecture, run the PyHealth Trainer, then evaluate with metrics that match the clinical question. Bypassing those stages usually fights the framework; following them keeps experiments composable and easier to hand off between agents. Use it during Validate when you need a credible prototype on public benchmarks, during Build when you implement the full pipeline, and during Ship when you tighten training and metric checks before release.521installs71ScanpyScanpy is an agent skill packaged as a complete single-cell RNA-seq analysis template for solo researchers and indie bioinformatics builders who need a reproducible Python path from raw counts to clusters without rewriting boilerplate each project. You point INPUT_FILE at an h5ad, tune MIN_GENES, MT_THRESHOLD, N_PCS, and LEIDEN_RESOLUTION, and run through load, QC, normalization, dimensionality reduction, and clustering with scanpy’s conventional settings. Figures and tables land under configurable OUTPUT_DIR and FIGURES_DIR with autosave enabled. It suits Claude Code or Cursor sessions where the agent fills in downstream annotation steps after the scaffold runs. Use it when you are building analysis pipelines or validating a dataset before publication-style figures—not when you only need a one-off plot without the full preprocessing chain.521installs72DeeptoolsdeepTools Quick Reference is an agent skill for solo builders and small lab-adjacent teams running ChIP-seq, RNA-seq coverage, or comparative BAM workflows from the terminal. It condenses the most common deepTools invocations—normalized bigWig export, log2 treatment/control tracks, multi-sample correlation heatmaps, TSS-centered plotHeatmap pipelines, and plotFingerprint enrichment QC—with flags such as --normalizeUsing RPGC, --effectiveGenomeSize, --binSize, and --numberOfProcessors. The documented flow moves from QC through bamCoverage and bamCompare to matrix computation and plotting, which matches how practitioners actually ship figures and supplemental tracks. Complexity is advanced: you need indexed BAMs, sensible controls, and organism-matched genome sizes. Use the skill when your agent should not improvise shell syntax for bioinformatics CLIs. It does not replace experimental design, peak calling, or publication-grade methods sections—you still own library choice and reviewer-facing rationale.520installs73PufferlibPufferLib is an agent skill that teaches solo builders and small ML teams how to create high-performance reinforcement-learning environments using the PufferEnv API and the Ocean environment collection. It is aimed at developers who already ship Python agent experiments and need vectorized rollouts without copying arrays on every step. Use it when you are implementing a custom environment, wiring multi-agent layouts, or adopting one of the 20+ Ocean presets before you commit to a training stack. The skill walks through space definitions, reset and step contracts, buffer-backed in-place updates, and how those choices affect PyTorch training loops. It matters because naive Gym-style envs often bottleneck indie RL projects; PufferLib’s design targets throughput for research prototypes and production-ish training pipelines alike.520installs74AdaptyvAdaptyv is an agent skill that packages the Adaptyv Bio Foundry public API as procedural knowledge for solo and indie builders shipping scientific automation. It documents how to create and manage wet-lab experiments over HTTPS—from drafting ExperimentSpec payloads with affinity, screening, thermostability, fluorescence, or expression types through quotes, invoices, tokens, sequences, results, targets, updates, and feedback. The audience is developers and agent authors who already use Adaptyv Foundry and want Claude Code, Cursor, or Codex to emit correct request bodies, required fields such as method and target_id for binding runs, and lifecycle flags without spelunking scattered docs. Use it during Build when you are coding integrations, orchestration scripts, or agent tools that submit amino-acid sequences, poll results, and react to webhooks. It matters because lab APIs are strict about enums and UUIDs; a single wrong experiment_type or missing replicate block wastes cycles and money on failed orders.519installs75Benchling IntegrationBenchling Integration is a procedural reference for solo builders and small teams shipping software that must talk to Benchling’s cloud R&D platform. It concentrates on the REST API v2 surface: how to target https://{tenant}.benchling.com/api/v2, authenticate with an API key or bearer token, and interpret list versus single-entity payloads. The material is aimed at agent-assisted development—your assistant can generate correct curl snippets, header sets, and pagination loops instead of guessing tenant URLs or auth schemes. Use it when you are building LIMS connectors, sequence importers, automation around registry objects, or internal tools that sync bench data with your own backend. It does not replace Benchling’s full OpenAPI catalog; it bootstraps the repetitive integration decisions every project hits first: auth, versioning, and paging. Intermediate familiarity with REST and secrets handling is assumed.519installs76Pydeseq2pydeseq2 is an agent skill that packages PyDESeq2 API knowledge for differential expression analysis on RNA-seq count data. Indie researchers, bioinformatics-heavy founders, and small teams building analysis services use it when they need statistically sound DESeq2 steps in Python without re-reading the entire library docs. The skill explains how DeseqDataSet ties together sample-by-gene counts, metadata, and design formulas such as condition or multi-factor models, then what deseq2() executes from normalization through dispersion estimation, LFC fitting, and outlier handling. It targets builders who already have count matrices and experimental designs and want the agent to scaffold correct method calls, parallelization, and AnnData export. This is advanced work: mis-specified designs or count orientation break results. The skill does not replace experimental design review or raw QC—it makes agent-generated analysis code align with PyDESeq2’s intended pipeline.519installs77Zarr PythonZarr Python is a reference skill for solo builders and small teams shipping data-heavy products—research tooling, geospatial apps, or ML feature stores—that must persist arrays larger than RAM. It documents how to install Zarr on Python 3.11+, attach S3 or Google Cloud filesystem drivers, and create stores with explicit chunk and dtype choices for efficient partial reads and writes. The skill walks through core constructors such as create_array, zeros, ones, and full, emphasizing cloud-native workflows and compatibility with Dask and Xarray for lazy, distributed computation. Use it during Build when you wire ingestion, transformation, or serving layers that should not depend on a single monolithic NumPy file on disk. Intermediate complexity: you should understand chunking tradeoffs and basic object-store credentials. It is not a full MLOps platform—pair it with your orchestration and monitoring stack for production Operate concerns.517installs78DatamolDatamol is an agent skill that acts as a focused reference for the `datamol.conformers` module: how to embed 3D structures, tune embedding methods, minimize energy, and cluster redundant poses. It walks through `generate` parameters such as conformer count, RMS deduplication thresholds, hydrogen handling, random seeds for reproducibility, and clustering with RMS cutoffs. Builders working on computational chemistry, molecular property models, or internal research tools install it so coding agents emit correct datamol idioms instead of guessing RDKit flags. It assumes Python and the datamol stack—not a no-code pipeline. The readme is API-centric rather than an end-to-end drug-discovery playbook, which keeps the skill narrow and precise for backend implementers who already know the science context.516installs79Molecular DynamicsMolecular-dynamics is a procedural reference skill for scientific-agent workflows built around MDAnalysis in Python. Solo builders and small research teams use it when they already have MD trajectories or structures and need reliable selection syntax, Universe setup, and analysis-module entry points instead of digging through scattered docs mid-session. The skill documents how to load topologies with trajectories, introspect atom and residue counts, and express complex AtomGroups with MDAnalysis selection language—including proximity to ligands, residue ranges, and negated solvent masks. It also surfaces alignment and RMSD/RMSF analysis imports so an agent can scaffold reproducible scripts for stability and flexibility metrics. It does not replace domain expertise in force fields or simulation setup; it accelerates the analysis layer once data exists. Install it when your agent must write or extend trajectory analysis, reporting, or pipeline glue for structural biology or computational chemistry side projects.516installs80Cellxgene Censuscellxgene-census is an agent skill that packages a structured reference to the CZ CELLxGENE Census: a versioned single-cell atlas on TileDB-SOMA. Solo builders and small teams shipping bioinformatics or ML side projects use it when they need accurate field names, experiment layout, and Python access paths instead of guessing from scattered docs. The skill fits the research phase of a data-heavy product—prototypes, notebooks, or pipelines that pull human or mouse obs/var and RNA measurements. It matters because wrong joins or misunderstood layers silently skew counts and gene presence. Pair it with your own analysis code; it does not run queries by itself but steers the agent toward valid census_info summaries, dataset metadata, and organism-specific SOMAExperiment usage.515installs81DeepchemDeepChem is a reference-oriented agent skill that maps DeepChem’s Python APIs for scientific machine learning—especially molecular and biological data. Solo builders and small teams use it while implementing cheminformatics, toxicity prediction, or sequence-based models where raw documentation is sprawling. The skill organizes capabilities by concern: data loaders for CSV/SDF/images and FASTA/FASTQ/alignment formats; dataset classes for in-memory NumPy versus on-disk scale; and splitters for reproducible train/val/test design. It suits agents that need accurate class and method names when wiring training scripts, notebooks, or backend jobs—not a substitute for domain validation or wet-lab claims. Expect intermediate Python and ML literacy; pair with your project’s testing and data-governance practices before shipping production models.515installs82Depmapdepmap is an agent skill for solo builders and tiny biotech research stacks who need disciplined reading of DepMap gene dependency data without hand-waving CRISPR screen artifacts. It explains Chronos scores on the v5+ pipeline: how copy-number effects, guide efficiency, and growth rate enter the model, and how to read numeric bands from non-essential near zero down through pan-essential mediators around −1. The skill catalogs positive control gene families—ribosomal RPL/RPS, proteasome, spliceosome, replication, and transcription machinery—and contrasts them with non-essential references for QC. For drug-target thinking it includes a selectivity computation pattern over gene effect matrices by cancer lineage so agents do not confuse broad essentiality with tumor-selective vulnerability. Advanced complexity applies: you need DepMap extracts, sane gene symbols, and Python pandas fluency. Prism positions it for research agents in Idea and early Validate, not for production clinical decisions. Pair with your own statistical review and licensed data-use policies.515installs83Imaging Data CommonsImaging Data Commons BigQuery Guide is an agent skill for solo builders and small research teams who already use IDC but hit idc-index limits. It explains dataset layout, authentication with application-default login, and query patterns for cohort discovery, complex clinical joins, and imaging-specific tables such as segmentations and structured reports. Use it when you need vendor-specific tags, nested DICOM sequences, or radiomics-style measurements that the mini-index does not surface. For most downloads and simple lookups, the skill still points you back to idc-index as the default path. The guide is intermediate: you should understand SQL, GCP projects, and why billing enables queries beyond the free tier allowance.515installs84Scvi Toolsscvi-tools differential expression is an agent skill for solo builders and small bioinformatics teams who need reproducible DE testing on single-cell data without defaulting to bulk RNA shortcuts. It walks agents through scvi-tools’ probabilistic framework: sample cellular states from the trained model, generate expression under the generative process, aggregate to population means, and test contrasts with Bayesian machinery that respects batch structure and zero inflation. The skill emphasizes when DE on latent or denoised representations beats naive count tests, and how to compare modalities supported by the ecosystem. It fits builders shipping analysis notebooks, internal research agents, or validation pipelines where effect sizes need uncertainty, not only p-values. Complexity is advanced: you need annotated AnnData, trained scVI-family models, and clear contrast definitions. Prism lists it for agent-assisted omics workflows—not as a substitute for statistical review or clinical interpretation.515installs85ArboretoArboreto is an agent skill that teaches how to run gene regulatory network (GRN) inference with the Arboreto Python library’s GRNBoost2 and GENIE3 algorithms. Solo builders and small bioinformatics teams use it when they already have an expression matrix (DataFrame, ndarray, or sparse CSC) and a list of transcription factors and need ranked regulator–target hypotheses without hand-rolling regression code. The skill contrasts efficiency and method: GRNBoost2 is the flagship choice for large single-cell and high-observation datasets thanks to stochastic gradient boosting with early stopping, while GENIE3 remains the random-forest alternative when that tradeoff fits your workflow. Both follow the same inference loop—train per target gene, surface important features as regulators, emit importance scores—so outputs stay comparable across runs. It matters for agent-assisted science because it encodes library imports, parameter defaults, and selection criteria so coding agents do not guess APIs or pick the wrong algorithm for scale.514installs86Esmesm is a scientific-agent integration skill for EvolutionaryScale’s Python SDK as inference moves from Forge to the Biohub platform. Solo builders shipping protein structure features, bioinformatics agents, or notebook-to-production pipelines install it when they need all-atom ESMFold2 prediction without confusing outdated Forge-only docs. The reference centers authentication (ESM_API_KEY), reproducible installs from Biohub’s GitHub versus PyPI esm==3.2.3, and the forge client class that still names “Forge” while hitting Biohub for ESMFold2. It matters because mixing endpoints or package sources in one virtualenv silently breaks structure calls. Use when SKILL.md-style tasks point at biohub.ai or when you must choose ESMFold2 over legacy ESM3-only flows. Intermediate complexity reflects API keys, version pins, and client selection rather than front-end work.514installs87Fluidsimfluidsim is a scientific-agent reference for the FluidSim Python stack used in 2D/3D turbulence and related PDE simulations. Solo and indie builders who ship research agents, lab automation, or reproducible simulation pipelines install it when an agent must set params.forcing (tcrandom, proportional, or in_script), wire custom forcing in Fourier space, or replace default initialization without hand-waving API details. The SKILL.md emphasizes practical hooks—forcing_rate, correlated random forcing bands, and overriding compute_forcing_fft on sim.forcing.forcing_maker—so codegen stays aligned with how simulations actually run. It fits builders already committed to a numerical backend who need procedural knowledge agents can cite step-by-step, not a generic “run a sim” prompt. Complexity is advanced because correctness depends on wavenumber limits, energy injection, and operator arrays in spectral space.514installs88PysamPysam is a phase-specific agent skill from K-Dense scientific-agent-skills that teaches coding agents how to work with real sequencing artifacts—not abstract CSVs. Solo bioinformatics builders and indie lab software authors use it when implementing Python pipelines that must open BAM or CRAM alignments, stream variants from VCF/BCF, pull reference slices from FASTA, or chew through FASTQ QC steps with a consistent htslib-backed API. The skill emphasizes when to reach for pysam versus shelling out blindly: region fetches, pileup coverage, tabix-indexed queries, and variant iteration patterns appear as copy-ready snippets installed via uv pip install pysam. It fits Build/Integrations work on backend data jobs, HPC scripts, and reproducible analysis repos where mistakes in file modes or index assumptions corrupt downstream science. Pair it with your orchestration and testing practices; it does not replace domain variant-calling tools but standardizes how your agent writes the Python glue layer.514installs89AstropyAstropy is an agent skill focused on astropy.coordinates and the SkyCoord class so coding agents produce correct, unit-aware astronomy code instead of hand-rolled RA/Dec math. Solo builders creating sky viewers, observation planners, catalog matchers, or research automation benefit when agents must parse sexagesimal strings, build ICRS or galactic coordinate objects, and format positions for logs or UI tables. The readme emphasizes practical patterns: decimal degrees, hour-angle strings, galactic lon/lat, batch coordinate arrays, and accessing HMS/DMS components without losing dimensional correctness. Complexity is beginner to intermediate depending on whether transforms between frames and proper motion enter the session. It is a domain integration skill for Python scientific stacks, not a general web SEO or SaaS boilerplate helper. Pair it when your product touches FITS catalogs, telescope scheduling, or educational astronomy tools that must cite standard coordinate representations.513installs90Diffdockdiffdock is an agent skill that encodes how to prepare DiffDock custom inference runs: example CSVs pairing proteins and ligands (paths, SMILES, or sequences), plus a configuration template pointing at score and confidence model directories, EMA checkpoints, and stochastic sampling knobs. Solo builders in drug discovery, protein engineering, or academic ML who lack a dedicated cheminformatics teammate use it so agents generate valid batch inputs and sensible hyperparameters instead of hand-editing opaque config trees. The skill assumes you already have structures or sequences and GPU-friendly workdirs; it shines when you are automating repeatable docking studies or wrapping DiffDock behind a CLI or internal API. Complexity is advanced because mis-specified ligand formats or step counts silently waste GPU time. After runs, you typically visualize poses or feed hits into downstream filtering outside this skill.513installs91GenimlGeniml BEDspace is an agent skill for solo builders and small research teams who need StarSpace-style joint embeddings over genomic region sets and their experimental metadata. It targets workflows where BED intervals carry labels like cell type or tissue and you want similarity search that respects both sequence context and condition, not plain interval overlap. The documented path starts with preprocessing a folder of BED files against a universe reference, a metadata CSV with file_name alignment, and explicit label column selection, then continues into StarSpace training via geniml bedspace CLI commands. Use it when building genomics agents, batch annotation tools, or internal pipelines that must answer metadata-aware nearest-neighbor questions across large region libraries. Complexity is intermediate because you must curate universe files, metadata schema, and local StarSpace binaries. It is a procedural skill package for agent sessions, not a hosted embedding API.513installs92MedchemMedchem API Reference is an agent skill that maps medchem 2.0.5 for solo builders shipping cheminformatics features—virtual screening dashboards, compound QA microservices, or agent tools that gate molecules before expensive steps. It documents RuleFilters as the batch entry point: pass rule names or callables, run across sequences of SMILES or RDKit mols, and interpret boolean pass columns alongside optional molecular weight, cLogP, and TPSA descriptors. Individual functions in medchem.rules.basic_rules support single-molecule checks such as Lipinski Rule of Five, CNS-oriented rules, and beyond-Ro5 variants for large binding sites. Use the skill when your agent must call real medchem signatures instead of inventing filter logic. Intermediate complexity assumes pandas, RDKit/datamol ecosystems, and comfort interpreting pass_all versus pass_any semantics in production pipelines.513installs93Neurokit2Neurokit2 is a scientific-agent skill that documents NeuroKit2’s Bio module for multi-signal physiological integration. Solo builders and small teams shipping health, HCI, or research tooling use it when an agent must clean and feature-extract several body signals without chaining separate one-off scripts. The skill centers on bio_process(), which accepts optional ECG, respiratory, EDA, EMG, PPG, and EOG arrays and returns a unified signals table plus an info dictionary for peaks and parameters. It matters because wearable and lab pipelines often need synchronized columns—ECG_Rate beside EDA_Phasic and RSP_Rate—for downstream ML or reporting. Install it when your repo already uses Python and you want reproducible, agent-guided NeuroKit2 patterns instead of guessing column names from scattered docs.513installs94Pydicompydicom is a reference-oriented agent skill from the scientific-agent-skills collection that catalogs commonly used DICOM tags—patient identity, demographics, study instance UIDs, dates, and descriptions—with types and short clinical meanings. Solo and indie builders shipping research tools, radiology-adjacent SaaS, or internal imaging pipelines install it so coding agents do not guess tag numbers or confuse PN, LO, DA, and UI value representations. Use it during build when you are wiring pydicom readers, anonymization scripts, QC checks, or export to PACS-compatible structures. It does not run conversions by itself; it gives procedural knowledge the agent can cite while writing Python that touches DICOM headers. Expect intermediate familiarity with imaging concepts; the payoff is fewer silent metadata bugs and faster iteration when schema questions come up mid-implementation.513installs95RdkitRDKit is a comprehensive Python API reference skill for the open-source RDKit cheminformatics toolkit, aimed at solo builders and indie researchers who ship scientific agents, internal APIs, or CLI tools that manipulate small molecules and biopolymers. Install it when your agent must parse SMILES or SMARTS, load structures from MOL, MOL2, or PDB sources, or emit standardized representations without hallucinating function names or argument lists. The SKILL content mirrors how practitioners work in computational chemistry and ML-for-chemistry pipelines: start from a structure string or file, sanitize the molecule graph, then feed descriptors or conformers into models or databases. Because it is reference-shaped rather than a single runnable recipe, it pairs best with a repo that already depends on RDKit and needs consistent, copy-paste-correct calls during build and integration. Use it during backend implementation and data-pipeline wiring—not for product validation, marketing, or production monitoring—whenever SMILES-to-molecule conversion or file round-trips are on the critical path.513installs96Dhdna ProfilerDhdna-profiler is an agent skill for advanced reference on the DHDNA writing and cognition profiler. Solo builders use it when they need more than a generic “rewrite in my voice” prompt—they want structured reads on depth, emotion, strategy, and precision across samples. The skill maps domain-specific presets (academic, creative, business executive, technical docs, journaling) to which dimensions to weight and what linguistic cues to hunt for, plus typical communication topologies. It supports positioning copy in Validate, drafting in Build, and launch content in Grow without pretending one phase owns the methodology. Confidence stays high when SKILL.md triggers are explicit about analyzing existing text rather than generating greenfield specs. Pair it with editorial or planning skills when the output should drive a concrete artifact.512installs97PathmlPathml is an agent skill that teaches agents how to manage large pathology whole-slide workflows using PathML’s HDF5-centric data layer. Solo builders shipping computational pathology tools, internal research platforms, or agent-assisted batch converters need predictable storage for SVS-scale imagery, generated tiles, masks, features, and metadata without reinventing chunked I/O. The skill walks through loading SlideData from a slide file, generating tiles at a chosen pyramid level, running a preprocessing pipeline, and persisting results with to_hdf5 for one slide or an entire SlideDataset built from a glob of inputs. It also covers scaling across workers when datasets grow beyond laptop memory. Intermediate complexity reflects coupling tile geometry, pipeline definitions, and HDF5 schema expectations. Install it when your agent must implement durable pathology datasets for downstream sklearn or deep-learning trainers rather than one-off PNG exports.512installs98PymatgenPymatgen Analysis Modules Reference is an agent skill that teaches solo builders and small scientific teams how to use pymatgen for materials characterization and computational analysis without re-reading the entire upstream docs. It centers on practical snippets: building phase diagrams from entries, querying stable phases, measuring energy above hull, decomposing metastable compositions, and plotting results. Chemical potential diagrams and Pourbaix-style electrochemical stability workflows appear in the extended reference so agents can scaffold analysis scripts, notebooks, or API endpoints for alloy design, battery materials, or catalysis tooling. Install it when your product or internal agent needs deterministic pymatgen calls rather than hallucinated chemistry APIs. Intermediate complexity assumes comfort with Python, compositions, and DFT-style total energies as inputs.512installs99Pyopenmspyopenms is a scientific agent skill that teaches procedural use of the PyOpenMS mass spectrometry toolkit through Python bindings. It is aimed at solo builders and researchers shipping proteomics or metabolomics tooling—CLI utilities, batch processors, or backend services that ingest mzML and derive features. The captured guidance walks through MSExperiment lifecycle (load, enumerate spectra, filter MS2), MSSpectrum properties, and numpy peak extraction, emphasizing that objects are C++ cores exposed to Python. Use it during Build when integrating lab data pipelines, not for generic web CRUD. Expect advanced complexity: comfort with Python, spectral concepts, and local scientific dependencies. The skill is reference-dense integration knowledge rather than a one-shot generator; pair it with testing and data validation practices before production deploys.512installs