AutoClaw Launches Self-Evolution Mechanism and Skill Store

ME News reports that on April 17 (UTC+8), according to monitoring by Beating, Zhipu AI’s Agent framework, AutoClaw, has officially launched its self-evolution mechanism. This mechanism addresses the pain point of Agent “forgetfulness” by identifying user corrections, new methods, or expression preferences during conversations and converting them into permanent memory. Users no longer need to repeatedly instruct the Agent to “keep responses brief” or “avoid using dashes”—after being taught once, the Agent evolves autonomously. The self-evolution mechanism is triggered by keywords (e.g., “from now on,” “remember”) or automatic detection through experience extraction after complex tasks. After each conversation, AutoClaw scans the dialogue and presents an “Evolution Request” card for user approval before any memory is written. To avoid information noise, Zhipu has established high-quality evolution principles, favoring 1–3 deep evolutions per week over daily fragmented recordings. This transforms the Agent from a “passive instruction executor” into a system that continuously grows through use—the more users teach it, the more it aligns with their personal habits. Simultaneously launched is the AutoClaw Skill Store, featuring the first batch of Zhipu’s proprietary GLM Office Skills suite: five tools for PPT, DOCX, XLSX, PDF, and Charts. Powered by the GLM-5.1 model optimized for office scenarios, these skills support intelligent auto-formatting, cross-format conversion (e.g., Word to PDF), and synchronized multi-file generation. The store also introduces expert-co-created Skills, including the “Nüwa Skill” for replicating individual thinking patterns and the “Horizontal-Vertical Analysis Method” for in-depth research—each encapsulating industry-leading methodologies as plug-and-play Agent capabilities. (Source: BlockBeats)