Agent Memory plugin: what it installs and how to add it

Agent Memory gives Claude Code a four-tier memory ladder, from raw session transcripts to stable cross-project persona. L1 candidate memories are recalled when relevant, while L2 project context loads at session start and L3 persona is always available. Promotion requires repeated observations across sessions, days or projects, with redacted or contradictory claims blocked from automatic advancement.

Three hooks capture, redact, recall and promote memories in the background, and fail open if a hook cannot run. Use /cs:memory to inspect provenance, review contested claims, preview promotions, or explicitly adopt changes. Adoption is the only path that writes to CLAUDE.md, with backups made first.

What's inside

• 1 skill: agent-memory
• 1 slash command: /cs:memory
• 1 agent: cs-memory-curator
• 3 hooks