InternScience/Awesome-Scientific-Skills

What it solves

Thousands of Agent Skills for AI coding agents are scattered across official registries and community repositories. A researcher who needs to run a gene ontology enrichment, draft a grant proposal, and plot dose-response curves shouldn't have to dig through 30,000+ entries to find the right three skills, and then wire them together from scratch. This project solves that discovery and integration problem by gathering and organizing high-quality, science-relevant skills in one place.

How it works

The project is currently a curated link collection (Phase 1). It evaluates skills from two tiers of sources: official skill registries (like Anthropic, OpenAI, Hugging Face) and community curations (like awesome-openclaw-skills, awesome-claude-skills). It also highlights science-specific repositories for fields like bioinformatics, chemistry, academic writing, and literature search. All skills are mapped into a target taxonomy (wet-lab & life sciences, computational & data science, literature & knowledge, research operations, quantitative & financial research).

The curation follows strict criteria: scientific relevance, source quality, clear documentation (SKILL.md), open license (MIT, Apache 2.0), security (no undisclosed network calls), and no redundancy. The skills follow the open Agent Skills specification, making them compatible with Claude Code, Cursor, Codex, Gemini CLI, and other agents that support this standard. The repository also outlines future phases: Phase 2 will import selected skills into a unified directory structure; Phase 3 will create multi-skill workflows (composition recipes).

Who it’s for

This is for researchers in any scientific discipline — bioinformatics, chemistry, neuroscience, clinical research, data science, and more — who want to use AI agents to accelerate their work but don't want to spend hours hunting for the right tools. It also helps developers who build agent-based tools for scientific domains, by giving them a vetted list of skills to draw from or contribute to.

Highlights

• Science-first curation — every skill is evaluated for real research workflows, not marketing demos.
• Fully clonable — git clone once to get a working skill library; no account, registry, or API key needed.
• Cross-agent compatible — works with any agent that supports the open Agent Skills standard (Claude Code, Cursor, Codex, Gemini CLI, etc.).
• Open and auditable — all skill sources are public; review before you run.
• Safety-aware — includes a security notice and selection criteria that require open licenses and no hidden network calls.
• Active roadmap — plans for a community skill registry, skill testing in OpenClaw and Claude Code, and an automated skill creator from papers or agent traces.