Science Skills - Google DeepMind's open-source research skills toolkit

What are Science Skills?

Science Skills is an open-source collection of scientific research skills launched by Google DeepMind, designed to accelerate AI agent-driven scientific research workflows. The tool integrates over 30 life science databases and tools, including AlphaGenome, AlphaFold Database, and UniProt, covering fields such as genomics, structural biology, cheminformatics, and literature retrieval, enabling researchers to complete complex analytical tasks that would otherwise take hours in minutes.

Main functions of Science Skills

• Genomics analysis: Use techniques such as AlphaGenome for gene variant annotation, disease association analysis, and genome data mining.
• Structural biology researchIt integrates the AlphaFold Database, supporting tasks such as protein structure prediction, molecular docking, and structure alignment.
• Cheminformatics processingIt provides analytical capabilities related to compound property prediction, molecular screening, and drug discovery.
• Intelligent document retrieval: Automatically retrieve, filter, and synthesize academic literature using tools such as OpenAlex to generate structured reviews.
• Multi-database integrationUnified access to 30+ authoritative databases enables one-stop scientific querying and analysis across data sources.

The technical principles of Science Skills

• Standardized Skill ArchitectureScience Skills employs a unified three-tier modular protocol: each skill consists of... SKILL.md(Including YAML route frontmatter and Markdown execution directives)scripts/(Executable script tools) and references/(Supplementary documentation loaded on demand) constitutes a structure that makes skills reusable, scalable, and maintainable.
• Progressive disclosure and token efficiencyBy replacing traditional monolithic instructions with a layered architecture: the core module is always loaded (only 38.5% of the necessary content is retained), and auxiliary content (examples, backgrounds, templates) is dynamically loaded when the Agent explicitly needs it through tool calls, reducing the cost of a single call to Token by about 39% and achieving an end-to-end saving of 43.2%.
• Script execution isolation：scripts/ The code runs directly on the file system, and the execution result (such as verification status or value) rather than the code itself is injected into the Agent context. Compared with the traditional method of "generating code in the conversation and then executing it", it reduces token consumption by about 89% while ensuring execution determinism and reusability.
• Routing layer optimizationEach SKILL.md The top YAML frontmatter contains concise route descriptions for quickly matching user requests with corresponding skills. By compressing lengthy descriptions and removing non-route-related content (such as feature lists and trigger word enumerations), it ensures that only the necessary skill context is loaded, avoiding irrelevant information from consuming tokens.

How to use Science Skills

• Installation Skill PackInstall a complete skill set with a single click using the npx command:npx skills add google-deepmind/science-skills/
• Enable in AntigravityAfter downloading Google Antigravity, select the Science plugin in the "Build with Google" step; existing users can download the Science plugin in the settings.
• Configure environmentThe Agent will automatically install upon first triggering of the Science Skill. uv Package manager; it is recommended to restart Antigravity after installation.
• Configure API keySome skills, such as AlphaGenome and OpenAlex, require API keys. The agent will guide users to obtain and write these keys. ~/.env document.
• Natural Language CallAfter installation, you can directly ask the Agent scientific questions in natural language, and the Agent will automatically load the relevant skills and perform analysis.