Memory Architecture Library

Comprehensive patterns for AI agent memory management, context engineering, and session persistence.

When to Use

• Managing context windows critically
• Building agents that run across multiple sessions
• Implementing tiered memory systems
• Preventing context rot in long conversations
• Active retrieval patterns for large codebases

References

memory-architecture-core.md

Core 4-layer memory architecture (A/B/C/D) with compression-as-curation and verification.

context-compaction.md

Combat Context Rot by enforcing automatic compaction after extended conversations. Forces periodic context refresh.

context-compiler.md

Dedicated Context Compiler session to curate context packages for Worker agents, preventing bias from failed attempts.

active-retrieval.md

Progressive disclosure by actively retrieving only needed context. Grep for specific patterns instead of reading full files.

domain-memory-pattern.md

Two-agent system (Initializer/Worker) for multi-session projects with features.json and progress.md tracking.

Usage

Load specific references as needed:
view /mnt/skills/user/memory-architecture-library/references/context-compaction.md