Awesome Dynamic Agent Skills: A Curated List for LLM Agent Skill Systems

Introduction

The Awesome Dynamic Agent Skills repository is a meticulously curated reading list and a companion to the TMLR 2026 survey, "Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Libraries." This resource focuses on dynamic, self-evolving skill systems for Large Language Model (LLM) agents. It addresses the challenge of static skill libraries by presenting a framework where skill libraries are treated as learning objects that can grow, shrink, repair themselves, and even distill into weights.

The central thesis of the survey, and thus this repository, is that dynamic skill systems are lifecycle-managed, verified, evolving artifact stores for LLM agents. The collection audits 124 papers published between 2023 and 2026, offering a unified lens through which to understand this complex domain. Key components include a six-sense taxonomy of what constitutes a "skill," a seven-tuple skill-record schema, an eight-stage lifecycle for skill evolution, and a ten-operator vocabulary for skill manipulation.

Accessing the Repository

This repository serves as a comprehensive reading list, making it straightforward to access its valuable content.

To explore the curated papers and the detailed survey information, you can:

• Visit the GitHub repository directly: https://github.com/yubol-bobo/Awesome-Dynamic-Agent-Skills
• Access the dedicated project page for a more interactive experience: https://yubol-bobo.github.io/Awesome-Dynamic-Agent-Skills/

For those interested in contributing or browsing the content locally, you can clone the repository using Git:

git clone https://github.com/yubol-bobo/Awesome-Dynamic-Agent-Skills.git

Examples and Structure

The Awesome Dynamic Agent Skills repository is structured to provide a clear overview of the research landscape. The "Surveyed papers" section categorizes 124 papers into distinct areas, including:

• Foundational
• Skill-aware RL
• Lesson / heuristic memory
• Executable skill libraries
• Parametric / training-time
• Infrastructure & governance
• Benchmarks & evaluation
• Safety & audit
• Position papers & related surveys

Each paper entry follows a consistent format: Name · arXiv link · year · operators · one-line headline. For instance, Voyager · arXiv:2305.16291 · 2023 · A · Open-ended embodied agent; 3.3× items, 15.3× tech-tree progress over baselines.

Beyond the paper list, the repository details a "Seven empirical regularities" section, summarizing key findings from the surveyed literature. These regularities offer concise, evidence-backed insights into the behavior and effectiveness of dynamic skill systems, such as "Curated skills outperform unverified self-generated skills" and "Maintenance becomes load-bearing at scale."

Why Use This Resource?

This repository is an indispensable resource for anyone working with or interested in LLM agents and their evolving capabilities.

• Comprehensive Overview: It provides a broad and deep understanding of dynamic agent skills, covering a wide range of research from 2023 to 2026.
• Structured Knowledge: The unified taxonomy, lifecycle, and operator vocabulary offer a clear framework for analyzing and developing skill systems.
• Up-to-Date Research: With 124 audited papers, it ensures you have access to the latest advancements and empirical findings in the field.
• Practical Insights: The "Seven empirical regularities" distill complex research into actionable insights, highlighting what works and why in dynamic skill management.
• Facilitates Contribution: Clear guidelines are provided for community contributions, ensuring the list remains current and comprehensive.

Whether you are a researcher, a developer, or simply curious about the future of AI agents, this resource will significantly enhance your understanding of how LLMs can acquire, manage, and evolve their skills dynamically.

Links

• GitHub Repository: https://github.com/yubol-bobo/Awesome-Dynamic-Agent-Skills
• Project Page: https://yubol-bobo.github.io/Awesome-Dynamic-Agent-Skills/
• Paper (PDF): https://openreview.net/pdf?id=cjU3YbcRr8
• OpenReview: https://openreview.net/forum?id=cjU3YbcRr8
• Citation (BibTeX):

  @article{
  li2026dynamic,
  title={Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Libraries},
  author={Yubo Li},
  journal={Transactions on Machine Learning Research},
  issn={2835-8856},
  year={2026},
  url={https://openreview.net/forum?id=cjU3YbcRr8},
  note={Survey Certification}
  }
  

• Contributing: https://github.com/yubol-bobo/Awesome-Dynamic-Agent-Skills/issues (Issues page for contributions)