memory-doc

Claude Code memory management uses CLAUDE.md files and auto memory to persist project-specific instructions across sessions. It loads these files at session start, reading topic files on demand to maintain context for project, user, and organization scopes.

CLAUDE.md scopes define where persistent instructions apply: project scope targets a specific repository, user scope applies across personal sessions, and organization scope enforces rules for entire teams. They load hierarchically at session start based on their directory locations.

To set up auto memory, open the MEMORY.md file to view memory guidance, then run the /memory command. This lets Claude automatically write notes to capture patterns, corrections, and workflows within the .claude directory structure.

Yes, Claude Code supports path-scoped rules for targeted memory loading. The .claude directory organizes these rules and memory files, ensuring Claude only loads relevant instructions based on the specific file paths being accessed during a session.

CLAUDE.md files require frontmatter metadata to define their scope and loading behavior. This metadata specifies how memory storage locations interact with the .claude directory, determining the load order and applicability of rules across project, user, and organization levels.

Instructions fail to persist when CLAUDE.md files are missing or placed outside the correct memory storage locations. Ensure your files are properly formatted with frontmatter metadata and reside in the .claude directory structure for proper session-start loading.