AGENT SKILL

memory-kit

Persistent context management for Claude Code sessions. Save, load, update, share, and audit session memory via MEMORY.md. Prevents context loss on compaction or session restart. Use when starting a session, before compaction, syncing context across teammates, or pruning stale memory entries. Trigger with "save memory", "load memory", "memory audit", "memory share".

适用职业开发工程师
License 许可
MIT
最近更新
2026-06-24
Stars
★ 67
Forks
24

01 前言 Front Matter

name
memory-kit
tags
[ "productivity", "audit", "memory-kit" ]
author
builtbyzac
license
MIT
version
1.1.0
description
Persistent context management for Claude Code sessions. Save, load, update, share, and audit session memory via MEMORY.md. Prevents context loss on compaction or session restart. Use when starting a session, before compaction, syncing context across teammates, or pruning stale memory entries. Trigger with "save memory", "load memory", "memory audit", "memory share".
allowed-tools
Read, Write, Edit, Bash(git:*)
compatible-with
claude-code

03 SKILL.md 正文

Memory Kit

Current State

![ -f MEMORY.md ] && echo "MEMORY.md: $(wc -l < MEMORY.md) lines, last modified $(date -r MEMORY.md '+%Y-%m-%d %H:%M')" || echo "No MEMORY.md found" ![ -f tasks/current-task.md ] && echo "Active task file found" || echo "No task file"

Overview

Claude Code sessions lose context on compaction and restart. Memory Kit persists session state (goals, decisions, patterns, open questions) to a MEMORY.md file that survives across sessions.

Five commands cover the full lifecycle:

  • /memory-save — snapshot before compaction
  • /memory-load — restore at session start
  • /memory-update — log a decision mid-session
  • /memory-share — push to git for teammates
  • /memory-audit — prune stale entries

Prerequisites

  • A git repository (for /memory-share)
  • Write access to the project root (MEMORY.md lives there)

Instructions

  1. On session start — check for existing MEMORY.md in project root. If found, read and summarize the saved state. Ask the user whether to resume previous context or start fresh.
  2. On save (/memory-save) — scan the current conversation for goals, decisions, patterns, and open questions. Write a structured snapshot to MEMORY.md with timestamped sections.
  3. On update (/memory-update) — append the user's decision or note to the appropriate section in MEMORY.md without overwriting existing content.
  4. On share (/memory-share) — commit MEMORY.md and push to the remote branch so teammates can load the same context.
  5. On audit (/memory-audit) — review all entries in MEMORY.md, flag stale items (older than 7 days or referencing completed work), and prompt the user to confirm removal.

Output

The skill produces and maintains a MEMORY.md file containing:

  • Session metadata: Timestamp, branch, and project name
  • Goals: Current objectives carried across sessions
  • Decisions: Key choices made with rationale
  • Patterns: Recurring approaches or conventions discovered
  • Open questions: Unresolved items requiring future attention

Output Format

For the MEMORY.md template structure, see output-format.md.

Error Handling

For error scenarios and recovery behavior, see error-handling.md.

Examples

Save before compaction:

"Save my memory" → reads current context, writes snapshot to MEMORY.md

Load at session start:

"Load memory" → reads MEMORY.md, summarizes state, asks to resume or start new

Quick mid-session log:

"Log decision: using Postgres over SQLite for concurrent writes" → appends to Decisions section

Team sync:

"Share memory" → runs scripts/memory-share.sh, confirms push

Resources

04 同仓库更多 Skills