The /init command in Claude Code and similar agents generates CLAUDE.md or AGENTS.md files by scanning your codebase, but the result is often better used as a project exploration artifact than as an always-loaded memory file. The key question is whether each piece of information is worth loading before every single task. A practical keep/remove/move framework helps evaluate generated content: keep only durable, task-independent guidance; remove generic commands the agent can infer; move reusable workflows to skills and project-specific details to on-demand project docs. Avoid using @ imports in memory files unless you want those files auto-loaded every session. The ideal memory file is minimal and contains only guidance that prevents real mistakes in every session.
Table of contents
IntroWhat Belongs in a Memory FileKeep, Remove, or MoveArchitecture Snapshots Are Often Better ElsewhereSkills vs Project DocsBe Careful with @ ImportsPractical TakeawaysSummarySort: