mkurman 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
- Read the skill's
SKILL.mdfirst. 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. - Invoke Python only through
uv run. The deepmind scripts use PEP 723 inline# /// script ... # ///blocks, souv runwill resolve and install the right deps in an isolated cache (~/.cache/uv/). Never use barepython3orpip install. - Read each skill's Prerequisites before running. Most skills need
uvon PATH (zorai runtime has it), a~/.envfile with the relevant API key, and possibly a notification step that records the skill was used. - Never read,
cat,echo,printenv, oros.environ.getthe.envfile or its keys. Deepmind scripts load credentials viadotenvinside the script — they pull keys from disk but do not surface them in the agent's context. Keep it that way. - When running in zorai's sandbox, credential injection is handled by
the plugin settings system (see
zorai_plugin_sciencesub-plugins); for the long-tail skills here, usedotenv'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/ |
REST. |
literature-search-arxiv |
skills/literature_search_arxiv/ |
REST. |
literature-search-biorxiv |
skills/literature_search_biorxiv/ |
REST. |
literature-search-europepmc |
skills/literature_search_europepmc/ |
REST. |
literature-search-openalex |
skills/literature_search_openalex/ |
REST (key optional). |
ncbi-sequence-fetch |
skills/ncbi_sequence_fetch/ |
REST (EFetch). |
openfda-database |
skills/openfda_database/ |
REST. |
opentargets-database |
skills/opentargets_database/ |
REST (GraphQL). |
pdb-database |
skills/pdb_database/ |
REST. |
protein-sequence-msa |
skills/protein_sequence_msa/ |
Compute. |
protein-sequence-similarity-search |
skills/protein_sequence_similarity_search/ |
Compute. |
pubchem-database |
skills/pubchem_database/ |
REST. |
pubmed-database |
skills/pubmed_database/ |
REST. |
pymol |
skills/pymol/ |
Compute (PyMOL binary). |
quickgo-database |
skills/quickgo_database/ |
REST. |
reactome-database |
skills/reactome_database/ |
REST. |
string-database |
skills/string_database/ |
REST. |
ucsc-conservation-and-tfbs |
skills/ucsc_conservation_and_tfbs/ |
REST. |
unibind-database |
skills/unibind_database/ |
REST. |
uniprot-database |
skills/uniprot_database/ |
REST, also has a zorai sub-plugin. |
uv |
skills/uv/ |
Internal. Sets up uv. zorai runtime already provides it. |
scienceskillscommon |
skills/scienceskillscommon/ |
Internal. Shared helpers for other skills. Do not invoke directly. |
workflow-skill-creator |
skills/workflow_skill_creator/ |
Meta-skill. Skip; the agent does not need to author new skills mid-task. |
Shared runtime rules
- All scripts expect to be run from the skill directory (or with
--project $SKILL_DIRfor ad-hoc). The skill's ownSKILL.mdwill 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.mdsays to record aLICENSE_NOTIFICATION.txtin 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.mdfor 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.txtfor the exact commit hash this bundle was vendored at. To refresh, re-vendor at a newer commit and update the pin.