Memory Management
Claude Code auto-memory 与 CLAUDE.md 的管理方法 — 判断值不值得记、记到哪一层(行内注释 / .claude/rules / memory / CLAUDE.md)、怎么写 frontmatter 与 description、何时 update 何时拆分、索引 MEMORY.md 的容量与 hub 化。用户说"记一下""新增/更新记忆""沉淀经验""复盘""改 memory schema""audit memory""记忆系统乱了"时使用;功能开发或长 debug 收尾有值得沉淀的结论时也用。
Install
npx skills add jau123/claude-memory-manager --skill memory-managementgit clone --depth 1 https://github.com/jau123/claude-memory-manager.git /tmp/claude-memory-manager
mkdir -p ~/.claude/skills
cp -r /tmp/claude-memory-manager/. ~/.claude/skills/memory-managementPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# Memory Management memory 装**事实**(踩过的坑、业界数据、项目状态),本 skill 装**方法**(记什么、记到哪、怎么写)。不要把 schema 规范抄进 memory 文件。 ## 快速路由 | 用户说 | 做 | |---|---| | "记一下"(模糊)/ "复盘" / "开发完了看看有什么值得记" | Mode 1 候选评估 | | "记一下今天那个 X 坑"(具体单条) | Mode 2 新建 | | "更新 / 修正某条 memory" | Mode 3 update(优先于新建) | | "改 schema / 加 type / 改分组" | Mode 4 结构变更 | | "记忆系统乱了 / audit" | 跑 `bash scripts/audit-memory.sh`;没有则从 `templates/audit-memory.template.sh` 复制 | | 新项目零记忆起步 | 见 `templates/bootstrap.md` | ## 六条规则 **1. 先判断值不值得记。** 同一个坑第二次踩 → 必记;对抗审查得出的非显然判断 → 记 Why;业界对照 / 实测数字 → 记;一次性 debug 修完没有沉淀价值 → 不记(commit message 够了)。**"很久没读"不是删除理由**,删的依据是主题已消失或被取代。 **2. 再判断记到哪(四级路由)。** 问一句:未来需要这条时,我是盯着某个代码位置,还是在做跨文件决策 / 排查某类报错? | 触发 | 去处 | |---|---| | 盯某一行/块,一句话讲得清 Why | 行内代码注释(写前先 grep,往往已有) | | 一组具体文件的多点契约 / 检查清单 | `.claude/rules/<x>.md` + `paths:` frontmatter(进 git,碰路径自动注入) | | 场景 / 报错 / 跨文件 / 平台与 SDK 行为 / 墓碑 / SOP / 经济推理 | memory(本系统) | | 每个 session 都要守的规则 | CLAUDE.md | 绑文件的 feedback 默认拆:内核进注释,memory 只留跨切原则 + 指针。拿不准就留 memory,宁可少迁。 ⚠️ rules 注入边界(官方文档,2026-10 核对): - `paths:` 规则在 Read / Write / Edit 匹配文件时注入(早期版本 Write 新建文件不注入)。 - `.claude/rules` 目录或规则文件若是**指向工作目录外的符号链接**,按外部 import 处理:要先批准外部 import,批准后也只加载**不带** `paths:` 的规则 —— 带 `paths:` 的规则静默失效(实测确认,连批准提示都不弹)。规则文件放在仓库里的真实目录,别软链到仓库外。 - 用户级 `~/.claude/rules/` 的 `paths:` 在早期版本被忽略(issue #21858,已关闭);要依赖它先实测。 - 新建或迁移规则后,Read 一个匹配文件确认规则真的注入了;注入不了的契约留在 CLAUDE.md。 **3. 优先 update 已有文件。** 主题相关就 update(加段或 changelog);只有主题完全正交才新建。碎片化 = 找不到 = 失忆。
Details
- Skill name
memory-management- Repository
- jau123/claude-memory-manager
- License
- MIT
- Repo updated
- 5 October 2026
- Listed in
- karanb192/awesome-claude-skills
FAQ
What does the Memory Management skill do?
Claude Code auto-memory 与 CLAUDE.md 的管理方法 — 判断值不值得记、记到哪一层(行内注释 / .claude/rules / memory / CLAUDE.md)、怎么写 frontmatter 与 description、何时 update 何时拆分、索引 MEMORY.md 的容量与 hub 化。用户说"记一下""新增/更新记忆""沉淀经验""复盘""改 memory schema""audit memory""记忆系统乱了"时使用;功能开发或长 debug 收尾有值得沉淀的结论时也用。
How do I install the Memory Management skill?
skills CLI (Claude Code, Cursor, Codex and more): npx skills add jau123/claude-memory-manager --skill memory-management. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is Memory Management popular?
It lives in jau123/claude-memory-manager, which has 4 GitHub stars and 0 forks, and is included in 1 curated skill list.
Related skills
Obsidian Skill
5gmickel/obsidian-skill
Use Obsidian's official CLI to work with a running Obsidian app and its configured vaults. Use for live workspace state, indexed search and metadata, daily notes, tasks, links and backlinks, Bases, templates, bookmarks, plugins, themes, Sync status, and developer inspection. Prefer ordinary filesystem tools when the task only needs direct file reads, writes, listing, or text search.
Job Application Tracker
5daliaabbruciati/job-application-tracker-SKILL
Automatically syncs Gmail and Google Calendar with a Notion database to keep track of the status of job applications the user has submitted. Use this skill whenever the user asks to "sync my job applications," "update my job tracker," "check where my interviews stand," or something similar — even if they don't explicitly mention Notion, Gmail, or Calendar, and even if they don't have a Notion database set up for this yet. Requires (or guides the user through setting up) the Notion, Gmail, and Google Calendar MCP connectors.
Task Decomposer
4sametcelikbicak/task-decomposer
Use when the user has a large, complex, or ambiguous request that needs to be broken down into smaller, actionable steps for AI assistance. Helps users structure prompts, manage context window limits, plan multi-session work, and get better results from AI coding agents. Also for 'too complex', 'where do I start', 'break this down', 'step by step', 'context too large', 'multi-step task', 'prompt structure', 'AI workflow', 'organize my request'.
Depth Lock
5malakhov-dmitrii/depth-lock
Lock scope, review rounds and wall-clock checkpoints before non-trivial work, and send everything outside the task to follow-ups. Use at the start of any plan, feature, fix train or multi-agent workflow, when a review loop starts repeating, when findings drift beyond the task, or when the operator says "не полируй", "не растягивай", "полировка круглого камня", "stop polishing" or "timebox".
Blueprint
5antbotlab/blueprint
Create a construction plan for a multi-step engineering task that requires multiple PRs or a phased approach. Produces a self-contained plan file with dependency graph, review gates, and step-by-step execution instructions for coding agents. TRIGGER when: user requests a plan, blueprint, or roadmap for a complex task, or describes work that clearly needs multiple PRs. DO NOT TRIGGER when: task is completable in a single PR or fewer than 3 tool calls, user asks to modify/update an existing plan, user says "just do it", or user asks for code review/audit.
Workflow Herdr
5letya999/workflow-herdr
Start a Herdr work run for a change. Use when starting work, doing a task in Herdr, splitting work across panes, or chatting from a Herdr terminal - even if they just say do this. Do NOT use for a solo one-pane edit, headless delegates, or product job queues.
Browse the most popular Claude skills, all productivity & notes skills, or write your own with the skills generator.