gdm-science-bundle

name

gdm-science-bundle

description

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`.

GDM Science Bundle (vendored)

A vendored copy of the google-deepmind/science-skills bundle (pin: see UPSTREAM_PIN.txt). This directory is the long-tail fallback: for the 5 sub-plugins with first-class zorai support (alphagenome, alphafold, uniprot, clinvar, chembl) prefer the matching plugins/zorai-plugin-science/ sub-plugin for typed settings and structured responses. For every other sub-skill in this bundle, follow the workflow below.

How to use a sub-skill in this bundle

1. Read the skill's SKILL.md first. Layout: skills/<skill_dir>/SKILL.md — the file's YAML frontmatter is the routing contract; the markdown body is the full workflow with steps, error fixes, and report templates. Never skip this read.
2. Invoke Python only through uv run. The deepmind scripts use PEP 723 inline # /// script ... # /// blocks, so uv run will resolve and install the right deps in an isolated cache (~/.cache/uv/). Never use bare python3 or pip install.
3. Read each skill's Prerequisites before running. Most skills need uv on PATH (zorai runtime has it), a ~/.env file with the relevant API key, and possibly a notification step that records the skill was used.
4. Never read, cat, echo, printenv, or os.environ.get the .env file or its keys. Deepmind scripts load credentials via dotenv inside the script — they pull keys from disk but do not surface them in the agent's context. Keep it that way.
5. When running in zorai's sandbox, credential injection is handled by the plugin settings system (see zorai_plugin_science sub-plugins); for the long-tail skills here, use dotenv's normal on-disk load.

Inventory of sub-skills in this bundle

Sub-skill (kebab-case)	Path under this bundle	Notes
alphafold-database-fetch-and-analyze	skills/alphafold_database_fetch_and_analyze/	Compute, also has a zorai sub-plugin.
alphagenome-single-variant-analysis	skills/alphagenome_single_variant_analysis/	Compute + API key, also has a zorai sub-plugin.
chembl-database	skills/chembl_database/	REST, also has a zorai sub-plugin.
clinical-trials-database	skills/clinical_trials_database/	REST.
clinvar-database	skills/clinvar_database/	REST, also has a zorai sub-plugin.
dbsnp-database	skills/dbsnp_database/	REST (NCBI EFetch).
embl-ebi-ols	skills/embl_ebi_ols/	REST (OLS4).
encode-ccres-database	skills/encode_ccres_database/	REST.
ensembl-database	skills/ensembl_database/	REST.
foldseek-structural-search	skills/foldseek_structural_search/	Compute (Foldseek binary).
gnomad-database	skills/gnomad_database/	REST.
gtex-database	skills/gtex_database/	REST.
human-protein-atlas-database	skills/human_protein_atlas_database/	REST.
interpro-database	skills/interpro_database/	REST.
jaspar-database	skills/jaspar_database/

Shared runtime rules

• All scripts expect to be run from the skill directory (or with --project $SKILL_DIR for ad-hoc). The skill's own SKILL.md will tell you which.
• Output artifacts should go under the user's working directory (or the path the user specifies). Do not pollute skills/.
• Confirm with the user before invoking anything with a real cost or rate limit (AlphaGenome API calls, OpenAlex bulk pulls, Foldseek server queries, etc.).
• If a sub-skill's SKILL.md says to record a LICENSE_NOTIFICATION.txt in the skill directory, skip the file write when running through zorai (the bundle is read-only inside the repo); instead, mention the upstream license URL to the user once per session.

License & attribution

• Code in this bundle — Apache License 2.0 (see LICENSE).
• Documentation in this bundle — Creative Commons Attribution 4.0 International (CC-BY-4.0).
• Individual database providers have their own terms. See SKILL_LICENSES.md for the full list. You are responsible for ensuring that any data retrieved through these skills is used in compliance with the upstream provider's terms.
• Upstream repo: https://github.com/google-deepmind/science-skills
• Pin: see UPSTREAM_PIN.txt for the exact commit hash this bundle was vendored at. To refresh, re-vendor at a newer commit and update the pin.