Tag
This paper introduces EvoClawBench, a benchmark designed to test whether AI agents can learn reusable skills from their own execution runs. Experiments with multiple agent runtimes show that skill learning is selective and cost-sensitive, not an automatic benefit.
Hermes Agent's /learn feature allows users to feed it any design and extract design tokens, layout structure, interaction states, and component patterns, turning it into a reusable skill for consistent design output.
The article discusses the growing adoption of SKILL.md for defining reusable agent skills, and questions its advantages over relying solely on AI tools like ChatGPT and Claude, considering factors like offline usage, standardization, workflows, and cost savings.
This paper argues that agent skills should incorporate visual information, not just text, and proposes a multimodal skill paradigm combining textual logic with visual support. Experiments show visual skills outperform text-only approaches in visual-centric tasks.
This paper introduces CARL, a method for offline hierarchical reinforcement learning that exploits local dynamics regularity to learn reusable skills. The approach clusters state-goal pairs requiring similar action sequences, enabling more effective skill reuse and improved performance on complex humanoid tasks.
Agent-Sin is an AI agent that automates repeated tasks using reusable skills, aimed at boosting productivity.