claude-code-memory-architect

Designing a Claude Code memory system across sessions involves an interview-driven process that selects building blocks like decay, promotion, multi-signal retrieval, salience, and compaction to build a personalized memory stack.

In an agent memory architecture, memory compaction reduces stored context to save space, while salience scoring ranks memory importance to prioritize retrieval during multi-signal retrieval workflows.

Wiring a memory recipe into CLAUDE.md or hooks requires completing the phased building workflow, which configures memory delivery as inline content, hooked events, or per-agent contexts based on your scope and requirements.

Building a custom memory stack without manual configuration is possible via an interview-driven design workflow that generates a memory recipe and handles end-to-end building with verification steps tailored to your fingerprint.

Memory decay and promotion are used in a multi-signal retrieval system to phase out stale context and elevate frequently accessed information, keeping the agent memory stack efficient across varying contexts and agent scopes.

Inline memory delivery risks consuming token limits quickly, whereas per-agent contexts offer scoped retrieval and compaction but require more complex hook wiring and multi-signal configuration during the phased build process.