EvoSkills provides a framework for extending AI agents, particularly EvoScientist, with specialized, installable knowledge packs called "skills." It aims to enhance AI capabilities for end-to-end scientific discovery by offering domain-specific expertise and enabling persistent learning across research cycles. The target audience includes users of EvoScientist and other coding agents seeking to augment their AI's research, experimentation, and writing workflows, ultimately amplifying AI potential in scientific endeavors.

How It Works

EvoSkills are modular, installable knowledge packs designed to integrate seamlessly with EvoScientist's multi-agent architecture (Researcher, Engineer, Evolution Manager) and its persistent memory system (evo-memory). This approach allows AI agents to acquire domain-specific expertise, enabling them to perform complex scientific tasks. Skills can be used individually or combined into pipelines, such as the research, experiment, and writing phases, facilitating a cohesive and evolving AI research process. The evo-memory component is central, enabling skills to learn and adapt across multiple research cycles, thereby improving future performance and direction.

Quick Start & Requirements

• Primary Install/Run Command:
  • Within EvoScientist: /install-skill EvoScientist/EvoSkills@skills
  • For other agents: npx skills add EvoScientist/EvoSkills
• Prerequisites: Compatible with "any coding agent." No specific hardware or software dependencies are detailed beyond the agent itself.
• Links: Installation commands serve as the primary starting point.

Highlighted Details

• Comprehensive Skill Suite: Offers over a dozen installable skills covering the full research lifecycle, including literature grounding (research-ideation), experiment execution (experiment-pipeline), iterative coding (experiment-iterative-coder), and end-to-end paper writing (paper-writing).
• Self-Evolution Loop: Core skills (research-ideation, experiment-pipeline, evo-memory) form a persistent learning mechanism that refines research directions and experimental strategies across cycles.
• MCP Server Marketplace: Provides a curated collection of servers to extend agents with external tools like web search and academic paper retrieval.
• AI-Generated Content: The nano-banana skill leverages Gemini for AI-generated presentation slides and illustrations, featuring an interactive review loop.
• Structured Workflows: Skills like paper-planning and experiment-pipeline enforce structured, multi-stage processes for research tasks, incorporating attempt budgets and gate conditions.

Maintenance & Community

The project is contributed to by Xi Zhang, Yougang Lyu, Dinos Papakostas, Yuyue Zhao, Xiaoyi, the DeepResearch Team, and the wider open-source community. Contributions are welcomed. For inquiries, contact EvoScientist.ai@gmail.com.

Licensing & Compatibility

This project is licensed under the Apache License 2.0, which is generally permissive for commercial use and integration into closed-source projects. The skills are designed to be compatible with "any coding agent."

Limitations & Caveats

Several advanced features are listed under "Coming soon," including dedicated modules for Math Olympiad, paper reproduction, grant writing, peer debate, trend analysis, and interactive paper QA. These indicate current gaps in the project's capabilities. The effectiveness of the skills is dependent on the underlying capabilities of the host coding agent.