计算机行业点评：再一次谈自定义Agent

Investment Rating - The industry is rated as "Buy" with an expectation of an increase exceeding 15% over the next 3-6 months [28]. Core Insights - The introduction of Agent Skills by Anthropic marks a significant breakthrough in the configurability of Agents, allowing for modular packaging of task instructions, code capabilities, and resource modules, thus enhancing reusability and configurability [10][11]. - OpenClaw, a new open-source Agent framework, provides a complete operational system that enables Agents to execute complex tasks continuously, distinguishing itself from traditional systems by incorporating a comprehensive execution chain [14][17]. - The development of custom Agents is shifting from a technology-driven approach to an application-driven one, with three key factors determining their capabilities: the native abilities of foundational large models, high-quality proprietary data resources, and developers' understanding of needs and scenarios [4][24][26]. Summary by Sections Skills: A Major Breakthrough in Agent Configurability - Agent Skills, launched by Anthropic, transforms Claude from a general conversational assistant into a specialized workflow executor, utilizing a structured folder system that serves as an AI "manual" [10][11]. - The core design philosophy of Skills is the "on-demand loading" mechanism, which optimizes context usage efficiency and enhances execution stability [11][12]. OpenClaw: Analysis of the Open-source Custom Agent Framework Mechanism - OpenClaw, launched in November 2025, is a fully open-source AI Agent framework that allows for extensive operational capabilities, enabling the execution of terminal commands and management of schedules through natural dialogue [14][17]. - It features a layered memory system and dynamic context management, allowing for continuous cognitive evolution and improved task consistency [17][20]. Determining Factors for Custom Agent Capabilities - The development threshold for custom Agents has significantly decreased, with the focus shifting to three main dimensions: the inherent capabilities of foundational large models, the quality and exclusivity of data resources, and the developers' ability to define needs and scenarios [4][24][26]. - The trend indicates that the ability to effectively define problems will be crucial for constructing viable solutions in the evolving landscape of AI [4][26].