defragmenting-memory

You can defragment memory by decomposing monolithic blocks into a hierarchical set of focused components using YAML frontmatter metadata. The process applies hierarchical block naming and a change-report workflow that creates new blocks, deletes obsolete ones, and syncs with memfs.

Memory block decomposition is the process of breaking down redundant or outdated agent memory into single-purpose components. Apply it when large memory blocks contain multiple topics, and when clear boundaries between identity, projects, and human context are needed to improve recall and safety.

Reorganize agent memory by applying hierarchical block naming and YAML frontmatter metadata to separate identity, projects, and human context. This structured decomposition creates clear boundaries that directly improve data retrieval recall and operational safety.

Yes, memory defragmentation supports a guided workflow to edit memory blocks in-place with memfs sync. This enables safety backups during the reorganization process while maintaining synchronization with the memory file system.

The best way to structure hierarchical memory blocks is to decompose monolithic data into focused categories like persona/identity, project/overview, and human/preferences. This single-purpose structure uses YAML frontmatter and produces a human-readable reorganization report detailing changes.

Agent memory retrieval fails or degrades when blocks contain mixed topics because large, multi-topic data lacks clear boundaries. Decomposing these monolithic blocks into focused, single-purpose components with hierarchical naming applies clear boundaries that improve recall.