Tag
EconSkills introduces a skill library and evaluation framework for web agents to transfer and retrieve procedural knowledge for live economic data retrieval, showing improved efficiency in controlled transfer and competitive performance at library scale.
Google Research's WikiSkill preprint demonstrates that compiling execution history into skill files allows a 9B AI model to outperform a 27B model in benchmarks, highlighting the role of procedural memory and skill transfer in agent performance.
WikiSkill uses a persistent wiki to evolve and transfer skills, showing that smaller models with such skills can outperform larger models without them in benchmarks, raising questions about what knowledge should persist in AI agents.
A thread explains that users can convert Opus 4.8 into Fable 5 for free before July 7, after which Fable 5 will cost credits. It provides tips for maximizing the remaining window, including having Fable 5 write a SKILL.md to transfer reasoning patterns to Opus 4.8.
Proposes a blind visual paradigm using Three.js to test if procedural scaffolds extracted from large models can improve small model outputs without fine-tuning, validated by a blind judge model.
Introduces AFTER, a benchmark of 382 enterprise tasks to evaluate procedural memory in LLM agents, showing that skill transfer improves performance across tasks, roles, and model backbones, with some skills generalizing broadly while others specialize.
This paper introduces SkillMigrator, an agent that learns reusable web skills as transferable interaction patterns (TIPs) and transfers them across websites by matching layout structure, reducing LLM action counts by 8-10% on benchmarks.
Hivemind is an open-source tool that allows multiple AI Coding Agents (e.g., Claude Code, Codex, Cursor) to share a memory layer, automatically mining high-quality patterns from usage trajectories and converting them into reusable skill files, enabling cross-tool and cross-team skill propagation, significantly reducing token consumption and interaction rounds.
LatentSkill converts textual skills into LoRA adapters stored in weight space, reducing context overhead while maintaining modularity and composability for LLM agents, achieving significant improvements on ALFWorld and Search-QA benchmarks.