google-deepmind/science-skills

Science Skills – AI‑agent extensions for scientific research

What it is – A curated bundle of agent skills that let a large‑language‑model‑based assistant (Google Antigravity) perform specialised scientific tasks. Each skill packages the prompts, helper scripts and reference data needed for a concrete workflow such as genomics analysis, structural‑biology modelling, cheminformatics queries, or literature search.

How it works – A skill lives in its own folder and contains:

• SKILL.md – a markdown file with YAML front‑matter that describes the skill (name, description, required APIs, parameters) and the step‑by‑step instructions the agent should follow.
• scripts/ – small utilities (Python, Bash, etc.) that the agent can invoke to run calculations, call external services, or transform data.
• references/ – optional PDFs, data files or links that the skill may need.

When an Antigravity user asks a scientific question, the agent selects the appropriate skill, installs any needed third‑party dependencies with the uv package manager, and runs the scripts automatically. If a skill needs an API key (e.g., AlphaGenome, OpenAlex, ClinVar) the agent prompts the user to add it to ~/.env.

Installation – One‑liner via the skills.sh ecosystem:

npx skills add google-deepmind/science-skills/

The command registers the whole bundle with Antigravity. Existing Antigravity users can also enable the Science plugin from the Settings → Customizations UI.

Typical use‑cases

• Genomics – retrieve gene‑variant data from AlphaGenome, filter by clinical significance, and output a summary table.
• Structural biology – fetch protein structures, run a lightweight folding script, and generate a visualisation.
• Cheminformatics – query chemical databases, compute molecular descriptors, and suggest synthesis routes.
• Literature search – call the OpenAlex API, rank papers by relevance, and produce a short annotated bibliography.

Extending / customizing – Users should not edit the files installed inside Antigravity’s plugin directory (they are overwritten on update). Instead, copy a skill to ~/.gemini/config/skills/ and modify it there, or create a new skill following the same folder layout.

Licensing – Code is Apache 2.0; documentation and non‑code assets are CC‑BY 4.0. Some third‑party data sources referenced by individual skills have their own terms, listed in SKILL_LICENSES.md.

Status – The repository includes a technical report describing the design and reliability considerations for the skills, and a demo page showing end‑to‑end workflows.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.