---
id: b159c0de-2e5e-4baa-a29b-0ab98f35cdbc
type: workflow
title: "/save-memories command + memory-saver agent for session memory capture"
tags: [cognitive-memory, claude-code, commands, agents, workflow]
importance: 0.7
confidence: 0.8
created: "2026-02-19T22:09:55.179468+00:00"
updated: "2026-02-28T20:57:20.251349+00:00"
relations:
  - target: c28f0a03-9fce-4cb9-8c26-715450891a92
    type: RELATED_TO
    direction: incoming
    strength: 0.8
    edge_id: 2411f2b3-c8b4-4bdc-a17b-9af2f2b8817a
---

Created a two-part system for saving session learnings to cognitive memory.

## Components

**Command** (`~/.claude/commands/save-memories.md`):
- Runs inline with full conversation context
- Finds cutoff point (last memory_store call) to avoid duplicates
- Analyzes conversation after cutoff point
- Builds structured summary with all session items
- Spawns memory-saver agent in background via Skill tool

**Agent** (`~/.claude/agents/memory-saver.md`):
- Runs on haiku model
- Receives structured summary as prompt
- Stores each item via MCP tools (with CLI fallback)
- Handles mechanical storage work

## Workflow

1. User calls `/save-memories` at session end
2. Command analyzes full conversation context
3. Command creates structured summary of learnings
4. Command spawns agent in background with Task tool and `run_in_background=true`
5. User gets control back immediately
6. Agent processes items asynchronously

## Advantages

- Replaces manual end-of-session memory saving
- Command handles context-dependent analysis
- Agent handles time-consuming storage operations
- Clean separation of concerns (context vs. mechanics)
- User not blocked waiting for storage to complete