SkillSmith: Enhancing Locally Deployed Agents via Automatic Skill Construction and Evolution

2026-08-08

AI Summary

This work addresses the performance gap between local small language models and cloud-based large language models, which stems from the former’s limited environmental knowledge. To bridge this gap, the authors propose SkillSmith, a cloud-edge collaborative framework that automatically constructs transferable skills from exploratory interactions of cloud agents and continuously refines these skills through execution feedback from local deployment. This non-parametric approach enhances the capabilities of frozen local models without requiring expert annotations, enabling, for the first time, fully automated skill construction, evolution, and cross-model transfer. Experiments on AppWorld and WorkBench demonstrate that a locally deployed Qwen3.6-27B agent achieves performance on par with cloud LLMs, reducing the average number of actions from 36.1 to 9.9, with learned skills generalizing effectively to other small models without retraining.

Abstract

LLM-based agent frameworks now act as personal assistants for multi-step tasks. Existing agent frameworks such as OpenClaw commonly follow the Cloud Agent depolyment mode using closed-source cloud LLMs as backbone model, which may expose private user information and incur repeated LLM-calling costs. Local Agents address these deployment concerns by depolying frontier open-source SLMs on user-controlled devices, but their task effectiveness still lags far behind Cloud Agents. Through diagnostic analysis, we reveal that the limited effectiveness of Local Agents with frontier SLM backbones mainly comes from missing environment knowledge caused by limited backbone model scale including environment rules and operation procedures. To supply such knowledge non-parametrically, context-efficiently, and without expert authoring, we present SkillSmith, a Cloud--Local Agent collaboration framework that uses Skill as a context-efficient knowledge carrier, automatic constructs Skill from Cloud Agent task exploration and evolves Skill using Local Agent execution feedback to enhance a frozen Local Agent. Experiments on daily agent task datasets AppWorld and WorkBench show that the automatically generated Skill enables the Local Agent with Qwen3.6-27B(SLM) to achieve task effectiveness comparable to Cloud Agents with frontier LLMs, outperform the strongest non-parametric baselines, reduce average actions per task from 36.1 to 9.9 on AppWorld-Normal, and generalize to other SLM backbone models without rerunning Skill construction.

Problem

• Local Agents
• environment knowledge
• task effectiveness
• small language models
• cloud-agent gap

Innovation

• Skill Construction
• Local Agent Enhancement
• Cloud-Local Collaboration
• Non-parametric Knowledge
• Automatic Skill Evolution