Nielsen Norman Group 研究 Claude 重度用户如何管理上下文库
Context Libraries: How AI Power Users Manage Context Engineering
当前可读材料为文章导语片段。NN/g称，在其对Claude重度用户的研究中，参与者花在整理上下文库上的精力超过写提示词，并遇到文件过期或丢失的问题；片段未提供完整样本规模和研究方法。
Summary: Claude power users rarely edit their context files and often can’t find them. Your AI agent is only as powerful as its context library.
AI agents have made curating context as important as writing prompts. The agent can do the typing, but deciding what context to keep, where to put it, and when to update it is left entirely to the user — and even the experts are struggling. This is true for UX professionals as much as for the Claude power users from various industries in our recent study . Our participants spent more effort engineering context libraries than crafting prompts (most of which they casually dictated). Yet each had improvised a bespoke process, files went stale or missing, and many suspected that someone else was surely doing it better.
What Is a Context Library?
A context library is a collection of institutional and procedural knowledge, contained within markdown files or external software, that AI agents may reference and edit.
A context library does not live solely within markdown files, though this was certainly the prevailing file type used to store context by our participants. MCP and API connections expand the library to include external sources such as Notion databases, Slack histories, or Granola transcripts. Some context in the library is actively updated and maintained; some is merely referenced. The context within the library generally plays one of three roles (which we discuss in greater detail in another article ):
来源：Nielsen Norman Group · nngroup.com