google-deepmind/science-skills
36 skills48.2k installs93.8k starsGitHub
Install
npx skills add https://github.com/google-deepmind/science-skillsSkills in this repo
1Literature Search Arxivliterature-search-arxiv is an agent skill from google-deepmind/science-skills that apply literature-search-arxiv agent skill workflows from documented skill.md guidance. # arXiv Search and Retrieval ## Prerequisites 1. **`uv`**: Read the `uv` skill and follow its Setup instructions to ensure `uv` is installed and on PATH. 2. **User Notification**: If LICENSE_NOTIFICATION.txt does not already exist in this skill directory then (1) prominently notify the user to check the terms at https://info.arxiv.o Developers invoke literature-search-arxiv during idea/research work for documentation tasks. The skill documents triggers, prerequisites, and step-by-step workflows grounded in SKILL.md. Compatible with Claude Code, Cursor, and Codex agent runtimes that load marketplace skills. Review the Security Audits panel on this listing before installing in production environments.2kinstalls2Literature Search OpenalexThe literature-search-openalex skill is designed for query the OpenAlex scholarly database for research papers, authors, institutions, topics, sources, publishers, funders, geo-locations, and keywords. Use when searching academic. uv: Read the uv skill and follow its Setup instructions to ensure uv is installed and on PATH. .env file: Make sure the .env file exists in your home directory. Invoke when the user searching academic papers, resolving DOIs, downloading open-access PDFs, finding an author's publications, aggregating bibliometric data (citation counts, h-index, impact factor), exploring the research taxonomies, or performing DOI lookups.1.8kinstalls3Pubmed DatabaseThe pubmed-database skill >- # PubMed API ## Prerequisites 1. **`uv`**: Read the `uv` skill and follow its Setup instructions to ensure `uv` is installed and on PATH. **User Notification**: If LICENSE_NOTIFICATION.txt does not already exist in this skill directory then (1) prominently notify the user to check the terms at https://pubmed.ncbi.nlm.nih.gov/disclaimer/ and https://www.ncbi.nlm.nih.gov/home/about/policies/ and to always check the license of the papers retrieved by the skill for any restrictions, then (2) create the file recording the notification text and timestamp. **`.env` file**: Make sure the `.env` file exists in your home directory. Create one if it does not exist. **`NCBI_API_KEY`** (optional): Raises the NCBI E-utilities rate limit from 3 to 10 requests/second. The skill works without it, but a key is recommended if the user plans many queries or encounters a 429 error. The user can obtain one for free by registering at https://www.ncbi.nlm.nih.gov/account/settings/ 5.1.6kinstalls4Literature Search EuropepmcThe literature-search-europepmc skill > # Europe PMC Database A skill for searching, downloading, and exploring open-access papers from [Europe PMC](https://europepmc.org/) - a comprehensive, free life-science literature database with over 43 million abstracts and 9 million full-text articles. **`uv`**: Read the `uv` skill and follow its Setup instructions to ensure `uv` is installed and on PATH. **User Notification**: If LICENSE_NOTIFICATION.txt does not already exist in this skill directory then (1) prominently notify the user to check the terms at https://europepmc.org/ and to always check the license of the papers retrieved by the skill for any restrictions, then (2) create the file recording the notification text and timestamp. ## Core Rules - **Open Access Only**: This skill exclusively searches open-access content. The script automatically appends `OPEN_ACCESS:y` to every search query. Do NOT remove or override this filter. - **NEVER run python3 or python3 -c directly**: the system Python does not necessarily have all key dependencies.1.5kinstalls5Literature Search BiorxivThe literature-search-biorxiv skill bioRxiv and medRxiv Literature Search Prerequisites 1 uv Read the uv skill and follow its Setup instructions to ensure uv is installed and on PATH User Notification If LICENSE_NOTIFICATION txt does not already exist in this skill directory then 1 prominently notify the user to check the terms at https api biorxiv org and https www biorxiv org content about-biorxiv and to always check the license of the papers retrieved by the skill for any restrictions then 2 create the file recording the notification text and timestamp Search Strategy Guide Read First This skill browses a date-based preprint archive It is NOT a keyword search engine Choose your approach based on what you already know A DOI e g from a citation Use search_by_doi py Approximate date category Use search_by_dates py with a 1 4 week range and category Only a topic or keywords no date Do NOT use this skill for discovery Use a keyword-capable literature skill first to find relevant DOIs then return here to1.5kinstalls6Workflow Skill CreatorThe workflow-skill-creator skill > # Workflow-to-Skill Distiller Turns a completed workflow into a reusable agent skill. Specifically, this skill extracts patterns from an interaction or workflow that **already happened** and packages them. > [!CAUTION] **You MUST complete Phase 1 (Brainstorming) before writing any code > or SKILL.md content.** Skipping brainstorming produces skills that are either > too rigid or too vague. The brainstorming conversation is the most important > part of this process. ## Phase 1: Brainstorming (MANDATORY) Have an **iterative back-and-forth conversation** with the user. Do NOT ask all questions at once. Pick 2-3 relevant questions per round from the bank below, refine your understanding, and ask follow-ups. ### Round 1: Understand the Workflow Start by summarizing what you observed from the workflow, then ask: 1. "Here's my understanding of the workflow: [summary]. What would you change?" 2. "What are the expected inputs and outputs for this workflow?" 3. "How often do you expect to run this workflow?1.5kinstalls7Uvuv in google-deepmind/science-skills is a prerequisite skill that guarantees the fast uv Python package manager is available before any dependent Python-based AI coding tools run. The workflow first runs uv --version; if that succeeds, setup stops immediately. If uv is missing from PATH, the skill checks the default install location at $HOME/.local/bin/uv and installs uv when needed so downstream Science Skills CLI scripts can execute. Many skills in the Science Skills collection depend on uv being installed and reachable. Developers and agents reach for uv automatically when another skill requires uv or when Python package commands fail with command-not-found errors. The skill is narrowly scoped to environment readiness rather than project dependency resolution. Common triggers include missing uv, Science Skills Python script failures, and first-time setup of the science-skills toolchain on a fresh machine.1.4kinstalls8Clinical Trials Databaseclinical-trials-database is a reference skill from google-deepmind/science-skills that documents ClinicalTrials.gov REST API v2 querying mechanics for developers building research agents and health-tech prototypes. The skill covers five endpoints—/studies, /studies/{nctId}, /studies/metadata, /studies/enums, and /studies/search-areas—plus ranked text search parameters such as query.cond for condition filters and advanced expression syntax. Developers reach for clinical-trials-database when an agent must search paginated trial listings, fetch a study by NCT ID, or discover available metadata fields and enum values instead of guessing API shapes. Version details are not pinned in the manifest, but the source explicitly targets API v2.1.4kinstalls9Pdb DatabasePDB Database is a Google DeepMind science skill that downloads PDB coordinate files in mmCIF or PDB format from RCSB. The bundled script requires Python 3.10 or newer and fetches specific protein structures for scientific workflows or AI research prototypes. Developers reach for PDB Database when they need authoritative structural coordinates from the Protein Data Bank without writing custom RCSB download logic. The skill fits quick retrieval of named structures into local pipelines, notebooks, or model-training prep steps.1.3kinstalls10Alphafold Database Fetch And Analyzealphafold-database-fetch-and-analyze is a Google DeepMind science skill for retrieving AlphaFold Database structure predictions and analyzing Predicted Aligned Error (PAE) data to detect protein domain boundaries. The bundled Python workflow targets developers and researchers building structural bioinformatics pipelines who need programmatic access to predicted coordinates and confidence matrices rather than manual database browsing. The skill expects Python 3.10 or newer and focuses on PAE interpretation for domain segmentation decisions. Reach for alphafold-database-fetch-and-analyze when a pipeline must pull AlphaFold entries, inspect alignment error patterns, and derive domain boundary calls from prediction metadata. It complements broader protein-structure analysis work without replacing full molecular-dynamics or wet-lab validation steps.1.3kinstalls11Pubchem Databasepubchem-database is a science skill from google-deepmind/science-skills for querying NCBI PubChem through PUG-REST computed-property search and PUG-View record APIs. It documents URL paths across domains including compound, substance, assay, gene, protein, pathway, taxonomy, and cell, with namespaces for CID, chemical name, SMILES, and InChI identifiers. Developers use pubchem-database when building cheminformatics pipelines, drug-discovery agents, or lab tooling that must fetch structures, synonyms, and bioassay metadata without hand-crafting REST paths. The skill also references a pubchem_api.py wrapper and raw API fallbacks when wrapper coverage is insufficient.1.3kinstalls12Ncbi Sequence Fetchncbi-sequence-fetch is a bioinformatics skill from google-deepmind/science-skills that provides an NCBI E-utilities wrapper for protein and nucleotide sequence retrieval inside coding-agent workflows. The module exposes subcommands for core NCBI E-utility operations so agents can fetch canonical sequence records without hand-rolling HTTP calls to NCBI endpoints. Developers reach for ncbi-sequence-fetch when building computational biology pipelines, literature-linked sequence lookups, or agent tools that need FASTA-ready protein and nucleotide data from authoritative NCBI databases. The skill fits scientific automation where sequence accession lookups must stay reproducible and API-compliant.1.3kinstalls13Openfda Databaseopenfda-database is a science-skills reference for the openFDA REST API at https://api.fda.gov/{category}/{endpoint}.json. Without an API key, requests cap at 240 per minute and 1,000 per day per IP—limits an automated agent can exceed in one session—while keyed access allows 240 per minute and 120,000 per day. Authentication uses --api_key or FDA_API_KEY. The skill documents search syntax with field:term queries, AND combinations, and category endpoints for drugs, devices, and adverse events. Developers reach for openfda-database when building health-research agents, pharmacovigilance tooling, or compliance dashboards that need reliable FDA data access patterns.1.3kinstalls14Pymolpymol is an agent skill providing PyMOL quick-reference patterns for headless molecular visualization. Every script must start with a mandatory initialization sequence—import pymol, set pymol_argv to ["pymol", "-cq"] for command-line quiet mode, then call pymol.finish_launching() before importing cmd; reversing this order crashes the session. PyMOL runs with OSMesa software rendering—no GPU or X display available. Use cmd.png(path, width, height, dpi) for standard output; cmd.ray() works for ray-traced quality but is slow. cmd.draw() must never be used because it requires hardware OpenGL. Researchers and computational biologists reach for pymol when agents need to programmatically render PDB structures, generate publication figures, or batch-export molecular images in CI or headless server environments.1.3kinstalls15Uniprot Databaseuniprot-database is a skill from google-deepmind/science-skills that gives coding agents reliable access to UniProt protein records across UniProtKB, UniParc, and UniRef. Developers use it to search proteins, map identifiers, and pull functional annotations and publication-linked metadata without model guesswork. Prerequisites include installing uv per the bundled uv skill setup and notifying users about LICENSE_NOTIFICATION.txt when present. The skill explicitly excludes sequence alignment, protein folding, and similarity search, directing those tasks to specialized skills. Reach for uniprot-database when building proteomics tooling, annotation pipelines, or agent assistants that must cite real UniProt entries. Pair with clinvar-database when workflows span variants and protein context.1.3kinstalls16Clinvar Databaseclinvar-database is a Python client skill from google-deepmind/science-skills that lets developers query the NCBI ClinVar database through E-utilities directly from Claude Code, Cursor, or other agent sessions. The skill targets Python 3.10+ and wraps ClinVar lookups for pathogenicity, clinical significance, and variant metadata without hand-rolling NCBI API calls. Developers reach for clinvar-database when building pipelines, notebooks, or agent tools that must ground answers in authoritative variant records instead of model hallucination. It fits genomics microservices, research automation, and clinical informatics prototypes that need programmatic ClinVar access. The skill is narrowly scoped to ClinVar retrieval; pair it with broader science-skills for adjacent NCBI resources.1.3kinstalls17Chembl Databasechembl-database is a google-deepmind science-skills module for querying the ChEMBL database at https://www.ebi.ac.uk/chembl/api/data from Claude, Cursor, or Codex agents. It documents list, single, batch, and search endpoint patterns plus SMILES similarity search for molecules. Seven endpoints support free-text search—including activity, assay, molecule, and target—while standard JSON pagination uses limit and offset parameters. Developers reach for chembl-database when building cheminformatics features, drug-discovery tooling, or bioactivity lookups without hand-rolling ChEMBL API docs each time.1.3kinstalls18Protein Sequence Similarity Searchprotein-sequence-similarity-search is a Google DeepMind science skill that submits a protein sequence to the ColabFold MMseqs2 server and returns homologous sequences in a ranked similarity report. Licensed under Apache 2.0, the skill automates quick protein homologue search without requiring developers to manually configure MMseqs2 infrastructure. Developers reach for it when building bioinformatics pipelines, validating sequence hypotheses, or integrating homology lookup into agent-driven scientific workflows. The workflow centers on API submission, result download, and structured ranking of similar sequences from the query input.1.3kinstalls19Ensembl Databaseensembl-database is a science skill from google-deepmind/science-skills providing a concise reference for the Ensembl REST API at https://rest.ensembl.org. The skill documents GRCh38 and GRCh37 endpoints (https://grch37.rest.ensembl.org), content negotiation headers for JSON, plain text, FASTA, and GFF3 responses, and a 15 requests per second rate limit. Developers reach for ensembl-database when the bundled ensembl_api.py script does not cover a specific genomic query and custom REST calls are needed. Use it in Claude, Cursor, or custom agents that fetch gene annotations, sequences, or variant data from Ensembl. The reference supports building bioinformatics integrations without manually parsing the full Ensembl API documentation.1.3kinstalls20Opentargets Databaseopentargets-database is a Google DeepMind science-skills module for querying the Open Targets Platform GraphQL API, which aggregates multi-modal evidence from GWAS and eQTL genetics, pathways, animal models, and clinical trials to rank target-disease associations and surface known drugs. Developers run scripts/query_opentargets.py for API communication, response formatting, and automatic truncation of large payloads to save agent context tokens. Built-in subcommands include get-gwas-studies with an EFO disease identifier for federated genetic evidence linking variants to diseases and druggable targets. Reach for opentargets-database when bioinformatics or computational biology workflows need druggability scores, target rankings, or disease association evidence inside Claude or terminal sessions—not for general SQL database administration. The Python script pattern keeps repeated GraphQL queries reproducible and context-safe for long API responses during drug-discovery research pipelines in agent, notebook, or script sessions.1.3kinstalls21Protein Sequence Msaprotein-sequence-msa is a science skill from google-deepmind/science-skills for computing multiple sequence alignments (MSA) on protein sequences via the EBI Clustal Omega service. The bundled Python script requires Python 3.10 or later with scienceskillscommon and python-dotenv dependencies managed through uv. Developers reach for protein-sequence-msa when agent workflows or bioinformatics pipelines need aligned protein sequence output for homology analysis, structure prediction prep, or phylogenetic comparison. The skill wraps external MSA computation so Claude, Cursor, or custom agents can submit protein FASTA inputs and retrieve alignment results without manually configuring Clustal Omega API calls.1.3kinstalls22Embl Ebi OlsEMBL-EBI OLS is a science-skills reference from google-deepmind/science-skills that documents the OLS4 REST API at https://www.ebi.ac.uk/ols4/api for ontology search, autocomplete, and term retrieval. The skill catalogs key endpoints including /search, /suggest, /select, /ontologies, and /ontologies/{id}/terms with double-encoded IRI lookup patterns. Developers reach for EMBL-EBI OLS when bioinformatics scripts, annotation pipelines, or agent tools must resolve GO, EFO, or other controlled vocabularies without hand-maintaining term lists. The skill focuses on correct API usage for full-text search, ontology browsing, and class detail retrieval in life-science coding workflows.1.3kinstalls23Dbsnp Databasedbsnp-database is a google-deepmind science-skills module for querying NCBI dbSNP variant data through documented REST endpoints. It uses NCBI Variation Services at https://api.ncbi.nlm.nih.gov/variation/v0/, including /refsnp/{rsid} for full RefSNP JSON with variant_type values such as snv, del, ins, delins, and mnv, plus placement and allele data across assemblies. Developers reach for dbsnp-database when agents must resolve rsIDs, inspect genomic placements, or pull structured variant metadata without manually browsing NCBI portals during bioinformatics feature work.1.3kinstalls24Human Protein Atlas Databasehuman-protein-atlas-database is a reference skill from google-deepmind/science-skills for constructing case-insensitive HPA search queries using key-value filters such as protein_class, tissue expression, mRNA levels, and subcellular localization. The readme documents core syntax rules, including that spaced terms like protein_class:Transcription factors do not require quotes and that quoting can break queries. Developers reach for human-protein-atlas-database when building health, bioinformatics, or research agents that must retrieve structured protein records from the Human Protein Atlas instead of guessing filter field names. The skill focuses on query construction and field semantics rather than running a bundled CLI, making it a integration guide for agent workflows that call HPA search endpoints.1.3kinstalls25Reactome DatabaseReactome Database is a Claude Code skill that documents the Reactome AnalysisService API at https://reactome.org/AnalysisService for programmatic pathway analysis. It covers database name and version GET endpoints, single-identifier analysis with species and disease filters, Homo sapiens projection routes, and POST batch identifier analysis with pagination and sorting parameters. Developers reach for Reactome Database when agents or backend services must map gene or protein IDs to pathways, run enrichment from posted identifier lists, or embed Reactome results in drug-discovery and genomics pipelines without manually browsing the web UI.1.3kinstalls26String Databasestring-database is a google-deepmind/science-skills guide for functional and PPI enrichment using scripts/string_cli.py invoked with uv run. The enrichment command accepts protein identifiers and a species NCBI taxon ID, then writes TSV output with category, term, p_value, fdr, and description columns covering GO, KEGG, Pfam, InterPro, and SMART terms. A separate ppi-enrichment command tests whether a network has significantly more interactions than a background proteome. Developers reach for string-database when they need reproducible enrichment statistics from STRING inside automated research pipelines rather than manual web searches. Commands are designed for copy-paste terminal execution or agent orchestration with explicit output paths such as /tmp/enrichment.tsv.1.3kinstalls27Interpro Databaseinterpro-database is a Google DeepMind science skill that documents every InterPro REST API query parameter for use inside agent workflows via fetch_interpro_data and get_interpro_count helpers. Parameters follow the official InterPro Swagger specification at interpro7-swagger.yml, including a global page_size int that defaults to 20 with a maximum of 200. The skill notes that page_size=1 combined with get_interpro_count enables rapid bulk aggregations without downloading full result pages. Developers reach for interpro-database when building bioinformatics pipelines, protein annotation tools, or research agents that must query InterPro entries, domains, and families with validated pagination and filter parameters instead of guessing API shapes.1.3kinstalls28Foldseek Structural SearchFoldseek Structural Search is a google-deepmind/science-skills script skill that runs Foldseek search for a PDB or mmCIF file against structural databases directly from an agent workflow. The bundled script requires Python >=3.10 and executes Foldseek similarity queries as part of computational biology pipelines. Developers reach for Foldseek Structural Search when protein structure files need fast database scanning without manually chaining shell commands and parsing outputs. The skill targets structural bioinformatics engineers embedding Foldseek into automated research and annotation flows.1.3kinstalls29Alphagenome Single Variant Analysisalphagenome-single-variant-analysis is a DeepMind science skill that teaches coding agents to run single-variant genomic effect predictions with the pip package alphagenome. The workflow imports genome, track_data, transcript, ism, dna_client, variant_scorers, and plot_components, initializes a DNA client from a dotenv-loaded API key, and produces pandas DataFrames plus matplotlib plots for variant interpretation. Developers reach for alphagenome-single-variant-analysis when they need in silico variant scoring, in silico mutagenesis (ISM), or locus visualization without hand-rolling AlphaGenome API calls. The skill targets bioinformatics and computational biology engineers embedding AlphaGenome into reproducible agent pipelines.1.3kinstalls30Quickgo Databasequickgo-database is a Gene Ontology annotations reference skill from google-deepmind/science-skills that drives the QuickGO annotation subcommand via quickgo_tool.py. Developers search GO annotations linking gene products to Biological Processes, Molecular Functions, and Cellular Components using filters like UniProtKB geneProductId, taxonId 9606, evidenceCode ECO:0000269, and GO term IDs. Example commands export JSON with --limit 50 through uv run scripts/quickgo_tool.py annotation search. The skill fits computational biology pipelines where functional enrichment, evidence filtering, and GO term lookup must be reproducible from the command line rather than manual web UI queries.1.3kinstalls31Jaspar Databasejaspar-database is a Google DeepMind science-skills wrapper around the JASPAR API for AI coding agents. The skill ships as a Python ≥3.10 script module that depends on scienceskillscommon and exposes transcription factor binding profile queries without leaving the agent session. Developers reach for jaspar-database when building or extending genomics, regulatory genomics, or motif-analysis code that must fetch curated PWM/profile data from JASPAR instead of hard-coding matrices or scraping the web UI. The README positions it explicitly as a JASPAR API skill wrapper under Apache 2.0, making it suitable for reproducible bioinformatics automation inside Claude Code, Cursor, or Codex workflows.1.2kinstalls32Gtex Databasegtex-database is a Google DeepMind science-skills module that wraps GTEx Portal API V2 for baseline RNA expression and expression quantitative trait loci. The gtex_cli.py script exposes subcommands to resolve gene symbols to versioned GENCODE IDs, fetch median TPM expression, list top-expressed tissues, retrieve gene-level eQTLs, and search eQTLs in chromosomal regions—with pagination capped at 250 items per page and sequential fetching per GTEx Terms of Use. First use requires license notification to .licenses/gtex_database_LICENSE.txt. Outputs are JSON files under /tmp/ for agent consumption. Reach for gtex-database when you need normal-tissue expression atlases or variant-gene regulatory context—not protein-level expression, diseased tissue samples, or fetal expression data.1.2kinstalls33Encode Ccres DatabaseEncode-ccres-database is a Google DeepMind science skill that exposes the ENCODE SCREEN GraphQL API at `https://factorbook.api.wenglab.org/graphql` for agent-driven genomic lookups. It documents core queries such as cCRESCREENSearch with arguments for genome assembly (grch38, mm10), genomic coordinate ranges, cCRE accessions, CTCF z-score rank filters, and cell-type biosample filters. Developers building regulatory-genomics pipelines, annotation tools, or research agents reach for this skill when they need structured cCRE SCREEN results without hand-writing GraphQL against the Factorbook schema. The skill pairs schema reference material with query patterns agents can execute directly in Claude or Cursor sessions.1.2kinstalls34Gnomad Databasegnomad-database is a Google DeepMind science-skills module for querying the Genome Aggregation Database without hitting gnomAD APIs directly. Three uv-run helper scripts cover variant frequency by variant ID or rsID, gene constraint metrics including pLI and LOEUF (oe_lof_upper), and region or gene variant search with consequence filters. The skill mandates license notification to .licenses/gnomad_database_LICENSE.txt on first use and always routes queries through the bundled wrappers to enforce API rate limits gracefully. Outputs land as JSON files such as variant_frequency.json or gene_constraint.json for downstream agent reasoning. Reach for gnomad-database when you need population allele frequencies, loss-of-function intolerance scores, or regional variant lists—not somatic cancer tracking (COSMIC) or raw sequencing reads (ENA).1.2kinstalls35Ucsc Conservation And Tfbsucsc-conservation-and-tfbs is a Google DeepMind science-skills package with 624 installs listed on skills.sh. The skill lets coding agents pull UCSC Genome Browser conservation tracks and transcription factor binding site (TFBS) data inside reproducible scientific workflows. Bioinformatics developers use it when variant interpretation, motif analysis, or comparative genomics needs authoritative track metadata without manually clicking the UCSC portal. The catalog entry ranks 8100 on skills.sh and ships from the google-deepmind/science-skills GitHub source for agent-driven genomic queries.1.2kinstalls36Unibind Databaseunibind-database is a google-deepmind/science-skills entry listed on skills.sh with 622 installs that lets agents query unified binding affinity data for proteins, ligands, and small molecules. The skill integrates scientific database access directly into agent workflows for structure-based research, compound screening context, and affinity lookups without manual portal navigation. Developers in computational biology, cheminformatics, or ML-for-science pipelines reach for unibind-database when agent tasks need authoritative binding measurements grounded in a consolidated UniBind dataset.1.2kinstalls