claude-memory-architect

To design a persistent memory system for Claude Code, you complete an interview phase and repo audit to build a tailored memory spec using building blocks like decay, promotion, salience, and compaction. This spec is deployed via CLAUDE.md entries or hooks to preserve useful context across sessions.

Memory compaction and decay in AI session context are building blocks for managing information lifecycle. Decay reduces the prominence of older context over time, while compaction summarizes and condenses memory to retain essential details without unwanted carryover.

You prevent unwanted context carryover in Claude memory by enforcing explicit TTLs, retention rules, and disclosure protocols within a tailored memory spec. These constraints ensure only safe and useful context is promoted across sessions.

Yes, you can deploy a Claude memory spec using SessionStart hooks. The design workflow supports injection strategies that output memory rules via hooks, CLAUDE.md entries, or agent-scoped files to maintain separation and resilience.

The best way to implement multi-signal retrieval for AI memory is to define explicit retrieval signals within a personalized memory spec. The design process audits open-source memory repos to inventory patterns and build a tailored retrieval strategy.

You need agent-scoped files to separate Claude memory across projects when you require strict context isolation. The memory spec design process supports agent-scoped memory deployment to maintain separation and resilience for distinct project workflows.