agent-memory-implementation

Restructure a bloated MEMORY.md by converting it into a lean two-layer memory system with a pointer-only index under 200 lines and on-demand topic files. This architecture improves agent recall and determinism by reducing context bloat.

The two-layer memory architecture uses MEMORY.md as a small pointer index and separates detailed knowledge into on-demand topic files. This structure ensures agents load only relevant knowledge, preventing chaotic and inconsistent memory indices.

Add frontmatter containing name, description, and type to your topic files to standardize AI memory management. Enforcing this frontmatter ensures topic files can be correctly audited and loaded on demand by the agent.

Yes, the lean two-layer memory system is fully compatible with Claude Code autoDream. It enforces a pointer-only MEMORY.md under 200 lines to maintain efficient knowledge loading and deterministic agent behavior.

Handle superseded or conflicting memories by deleting or correcting them in place without creating an archive directory. This maintains a single source of truth and prevents the memory index from becoming bloated with outdated content.

Audit your memory files when your project has a long MEMORY.md index, scattered topic files, and missing or vague frontmatter. This audit checks frontmatter correctness, line counts, and pointer integrity to reorganize memories effectively.