Google DeepMind releases Science Skills for AI research agents

Google DeepMind has released Science Skills, a modular toolkit that extends AI agents with research capabilities in genomics, structural biology, cheminformatics and literature search. The package runs inside Google Antigravity, the company's agent environment, and can be installed with a single command via npx or through the platform's settings interface.

Installation and dependency management

Installation follows two paths. New users tick the Science box in the Build with Google step during first launch. Existing users update to the latest version, then download the add-on through Settings > Customizations > Build with Google Plugins. Behind the scenes the system uses uv, a modern Python package manager, which is installed automatically the first time a skill runs. The agent prompts for confirmation, installs uv, and recommends restarting Antigravity to complete the process.

API keys and extensions

Some skills require an API key to function. AlphaGenome and OpenAlex will not work without one. ClinVar operates without a key but with lower rate limits. The agent is expected to guide the user through obtaining keys and writing them to ~/.env — for example, `ALPHAGENOME_API_KEY=your_actual_api_key`. Anyone who wants to adapt an existing skill or create a new one is instructed not to modify files inside the plugin directory, since changes there will be overwritten on the next update. Custom versions should be saved in the personal directory `~/.gemini/config/skills/`.

Licensing and limitations

Code is distributed under Apache 2.0, accompanying materials under CC-BY 4.0, and third-party data sources referenced inside skill files carry their own licences — details appear in the Skill Licences and Terms of Use file. Google stresses that everything is provided "AS IS" without warranty and explicitly notes that this is not an official Google product.

Documentation and examples

A technical report describing the architecture was published alongside the code. Usage examples are available at antigravity.google/use-cases/science. The modular structure — each skill contains a SKILL.md file with YAML frontmatter, a scripts folder and an optional references folder — is designed to allow extension without breaking compatibility in future updates.