zorai

Use when provisioning Prime Intellect GPU compute, managing pods/disks/sandboxes, running hosted RL training via prime lab, installing or publishing RL environments, or exposing local services via Prime Tunnel. Covers the `prime` CLI (PyPI: prime) for all Prime Intellect platform operations.

Shared Python package for Science Skills, currently containing http_client -- a unified HTTP client with rate limiting, retries, and exponential backoff. Not a standalone agent skill. Do not invoke directly.

Retrieve and analyze AlphaFold predicted structures for a protein. Use when the user provides a specific UniProt Accession ID and wants structural confidence metrics (pLDDT), domain boundary analysis, or disorder assessment. Do not use if the user only has a protein name, gene name, or amino acid sequence — ask for a UniProt ID first.

Query the ChEMBL database for bioactive molecules, drug targets, bioactivity data, approved drugs, and chemical structures. Use when the user asks about compounds, targets, IC50/Ki values, drug mechanisms, or structure searches.

Query ClinicalTrials.gov via APIv2. Use when you want to search for trials by condition, drug, location, status, or phase; retrieve trial details by NCT ID; check eligibility/inclusion criteria; count trials across conditions or time periods; identify a sponsor's trial portfolio; find recruiting trials for patient matching.

Use when needing clinical significance, pathogenicity classifications (e.g., Pathogenic, Benign, VUS), clinical evidence rationales, or finding "hard positive" benchmark controls for human genomic variants.

Use when you want to look up, map, and search for short genetic variants (SNPs, indels) in NCBI's dbSNP database. Resolves between rsIDs, genomic coordinates in VCF format, and HGVS strings. For an rsID, returns variant type, gene associations, clinical significance, allele frequencies, and genomic coordinates (GRCh38).

Query and search the EMBL-EBI Ontology Lookup Service (OLS) for biomedical ontology terms, definitions, and hierarchies across 250+ ontologies (e.g., GO, DOID, HP). Use when the user asks to search for terms, retrieve details, navigate hierarchies (parents, children, ancestors), look up properties and individuals, get autocomplete suggestions, or access ontology metadata and statistics.

Query the ENCODE Registry of cis-Regulatory Elements (cCREs) via the SCREEN GraphQL API, or make custom queries to the ENCODE Portal REST API for experiments and files (ChIP-seq peaks, etc.). Use when you want to query regulatory annotations or raw experimental data across human cell types.

Query the Ensembl database to resolve gene, transcript, and protein IDs, fetch genomic or protein sequences, retrieve gene structures (exons), and get variant consequence and effect predictions (VEP). Use this skill as a primary ID translator, genomic sequence database and variant effect prediction tool.

Performs 3D structural searches of proteins against various databases (PDB, AlphaFold, CATH, MGnify, etc.) using the Foldseek API. Use ONLY when the user provides a physical 3D coordinate file (.cif, .mmcif, or .pdb) and wants to find structurally similar proteins. Do NOT use if the user only provides a protein sequence, gene name, or UniProt ID.

Query the Genome Aggregation Database (gnomAD). Use when determining the rarity or allele frequency of specific genetic variants, retrieving gene constraint metrics (pLI, LOEUF) to assess loss-of-function intolerance, finding variants in a genomic region or gene, or querying structural variants. Don't use for analyzing individual patient genomes, tracking somatic mutations in cancer (use COSMIC), or requesting raw sequencing reads (use ENA).

Use when you want to retrieve quantitative RNA expression data and variant eQTL information from the GTEx (Genotype-Tissue Expression) Project across 54 non-diseased tissue sites.

Use when you want to retrieve semi-quantitative protein expression and spatial localisation data from the Human Protein Atlas (HPA).

Identify domains, families, and sites in proteins; find all proteins in a family or sharing a domain; explore species distribution for a domain; annotate genomes with protein families and GO terms. InterPro combines 14 databases (e.g., Pfam, CDD) into one searchable resource. InterPro-N significantly expands annotation and sequence coverage with deep learning. Includes domain architecture (IDA) search.

Query the JASPAR database for Transcription Factor (TF) binding profiles. Use when retrieving Position Frequency Matrices (PFMs) or Position Weight Matrices (PWMs) for specific TFs, resolving gene symbols to JASPAR Matrix IDs, or getting TF metadata. Supports multiple output formats (MEME, TRANSFAC, PFM, JASPAR, YAML).

Search for scientific papers, preprints, and publications on arXiv. Extract metadata, abstracts, and download full-text PDFs or HTML versions of papers. Use when the user asks to find research papers, literature, or specific arXiv IDs.

Browse, filter, and download life sciences, biology, and medical preprints from bioRxiv and medRxiv. Supports fetching paper metadata by DOI, and browsing by date range with category and keyword filters. Keyword filtering is local, so date ranges MUST be narrow (1-4 weeks) with a category to prevent timeouts.

Search Europe PMC for scientific literature and download open-access full texts and PDFs. Retrieve full-text XML/plain text by PMCID, get citation lists and bibliography.

Query the OpenAlex scholarly database for research papers, authors, institutions, topics, sources, publishers, funders, geo-locations, and keywords. Use when 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.

Retrieve protein and nucleotide sequences from NCBI databases using E-utilities. Supports direct accession lookup, CDS translation, gene+organism search, locus lookup, PubMed-linked sequences, patent protein extraction, and organism+length fallback search. Use when you need to fetch biological sequences by accession, gene name, locus tag, PubMed ID, or patent number.

Query, search, and download data from the openFDA API for drugs, devices, foods, tobacco, cosmetics, animal and veterinary products, substances, and transparency data. Use for FDA adverse events, recalls, labeling, approvals, shortages, 510(k) clearances, NDC lookups, and any FDA safety or regulatory data query across all 28 API endpoints.

Query Open Targets Platform for target-disease associations, drug target discovery, tractability/safety data, genetics/omics evidence, known drugs, for therapeutic target identification.

Use when you want to search for or download experimentally-determined 3D structures for biomolecules (proteins, nucleic acids, bound ligands). Supports searching by sequence similarity, structure similarity, chemical and other attributes. Also use to get metadata about biomolecular structure experiments.

Performs multiple sequence alignment of proteins with EBI Clustal Omega. Use when you need to align multiple sequences to assess similarity, domain conservation, or key residue conservation. Supports up to 4000 sequences and a maximum file size of 4 MB. Do not use to search for homologous proteins in a database (use MMseqs2, BLAST), align non-protein sequences (DNA, RNA), perform structural alignment (use Foldseek, PyMOL), or if you only have a single sequence.

Searches for homologous protein sequences using MMseqs2 (fast, default) or BLAST (comprehensive, fallback). Trigger this whenever the user provides a protein sequence or FASTA file and asks to find homologues, sequence matches, or wants to infer protein function based on sequence similarity, but not when the user wants to infer protein function based on structural similarity.

Query PubChem, search by name/CID/SMILES, retrieve properties, similarity/substructure searches, bioactivity, for cheminformatics. Use when a user asks about a specific chemical, drug, or molecule.

Search PubMed for scientific literature, including published clinical trials. Fetch abstracts and full text. Link published research to biological databases (gene, protein, nucleotide, PubChem) to discover associations between papers and specific compounds or genes. Verify medical spelling, match raw citations, and cache result sets for bulk processing. Interfaces NCBI E-utilities and PMC BioC APIs.

Query the QuickGO and Evidence & Conclusion Ontology (ECO) REST API. Use this when you need to map genes to biological processes, molecular functions, or cellular components, find genes associated with a specific pathway/GO term, or explore the Gene Ontology hierarchy. Do not use for querying drug targets (use OpenTargets) or mechanistic signaling pathway diagrams (use KEGG).

Query the Reactome database (Analysis and Content Services). Use when the user asks about pathway analysis, gene list enrichment, retrieving results by token, finding unmapped or not-found identifiers, mapping identifiers, reaction participants (inputs, outputs), pathway hierarchy (including top-level pathways), diagram export, cross-reference mapping, or searching the knowledgebase.

Vendor of the google-deepmind/science-skills bundle (37 skills for scientific research). Use when a user asks about any of: AlphaGenome single-variant effect analysis (RNA-seq / DNase / ChIP / TF effects, splicing disruption, UBERON/CL ontology resolution for non-coding variants), AlphaFold DB fetch and analyze, ChEMBL bioactivity queries, ClinicalTrials.gov lookups, ClinVar variant interpretation, dbSNP, EMBL-EBI Ontology Lookup Service (OLS4), ENCODE cCREs, Ensembl REST, Foldseek structural search, gnomAD, GTEx, Human Protein Atlas, InterPro, JASPAR transcription-factor profiles, literature search (arXiv / bioRxiv / EuropePMC / OpenAlex), NCBI sequence fetch (EFetch), openFDA, OpenTargets, PDB, protein sequence MSA / similarity search, PubChem, PubMed, PyMOL structural visualization, QuickGO, Reactome, STRING, UCSC conservation & TFBS, UniBind, UniProt, or any workflow combining them. Always read the per-skill SKILL.md under skills/<name>/ and invoke Python only through `uv run`.

Query the STRING database for protein-protein interactions (PPIs), functional enrichment, and homology. Use when the user asks about interactions between specific proteins, interaction evidence, confidence scores, protein interaction partners, or pathway enrichments.

Fetch Evolutionary Conservation scores (phyloP, phastCons) and Transcription Factor Binding Sites (TFBS) from the UCSC Genome Browser. Use when analyzing whether genomic variants or regions are evolutionarily conserved, functionally important, or bounded by TF regulators across major projects (ENCODE, JASPAR, ReMap).

Queries the UniBind database for experimentally validated transcription factor (TF) binding sites. Use when retrieving direct TF-DNA interaction datasets, downloading binding site coordinates (BED/FASTA) for local analysis, or listing available datasets by species, cell line, or TF name. Don't use to query specific intervals, locations, genes, motif models or expression data.

Access protein metadata, function, taxonomy, and sequences across UniProtKB, UniParc, and UniRef. Use when searching for proteins, mapping identifiers, or retrieving functional annotations and publications. Don't use for sequence alignment, protein folding, or sequence similarity search (use specialized skills for those tasks).

Distills a completed user workflow or interaction into a reusable agent skill. Use when the user asks to turn their workflow, interaction, or multi-step process into a skill, or when they say "make this a skill", "create a skill from what we just did", "package this workflow" or similar. Do not use for creating skills from scratch without an existing workflow (use a generic skill-creator for that).

Prepare and validate Parameter Golf record folders: self-contained train_gpt.py, README.md, submission.json, FineWeb SP1024 BPB accounting, artifact-size logging, run logs, and PR-ready folder hygiene.

Run Parameter Golf competition submissions on RunPod GPU Pods. Covers required operator inputs, RunPod pod specs, FineWeb SP1024 data caching, record-folder hygiene, torchrun launch commands, monitoring, artifact-size checks, and result collection.

Hands-on implementation template and API reference for writing, tuning, debugging, and benchmarking Triton GPU kernels. Covers the full triton.language API surface, autotuning patterns, profiling workflows, and production integration.

Hands-on implementation template and API reference for writing, tuning, debugging, and benchmarking Triton GPU kernels. Covers the full triton.language API surface, autotuning patterns, profiling workflows, and production integration.