disk-raccoon README

The project's README from github.com/orange2ai/disk-raccoon. As one markdown file: README.md.

硬盘小浣熊 🦝 Disk Raccoon

简体中文 | English

一个帮你清理 Mac 缓存的 AI skill。给你的 AI 编程助手装上它,它就能帮你找到硬盘里那些偷偷占了几十个 G 的缓存文件,判断哪些能安全删掉,然后帮你清理。

苹果的固态硬盘焊死在主板上,升一档 512G 加价 ¥1500。小浣熊第一次出手,一晚清出 135+ GB,零翻车。如果你的盘是被缓存塞满的,清一次就省了一笔升配钱。大概是你 agent 里最值钱的 skill,¥1500 起。

缘起

我用 Mole 扫了一下硬盘,找出几个很大的文件夹,每个都几十个 G。

删不删呢?

飞书那个目录,一半是文档缓存,删了没事;另一半是登录态,删了要重新扫码。Chrome 也一样,Service Worker 能删,旁边的 IndexedDB 删了网页应用数据就没了。

同一个文件夹里,一半是垃圾,一半是状态。扫帚能扫出大小,扫不出答案。

于是让 AI 来判断。它逐个去查每个目录属于什么应用、存的是什么、删了会怎样,查完列了一张单子给我。照着清,一晚清出 135+ GB,零翻车。

既然这么好用,我想到我还有一台电脑,不如把这套判断做成一个 skill,还能分享给朋友。就有了这个项目。

为什么叫浣熊

浣熊是自然界唯一吃东西前先把食物洗一洗的动物。

它不是不吃,是不瞎吃。

大多数清理工具扫全盘,把大的列出来让你勾。但 macOS 真正的问题是:大的不一定该删,该删的不一定大。

小浣熊的重点不是删,是分得清什么能删。 先洗,再吃。

只做缓存

卸载应用、管后台、优化系统,这些 Mole 🐹 已经做得很好了,开源免费。

小浣熊只管缓存,但管到底。核心是一份缓存地图(references/cache-map.md),每条路径标了安全等级:

地图上没有的路径?查清楚再说,查不清楚就不动。

安全设计

原则 说明
默认只读 第一轮只看不删,先出报告
先认清再删 地图上没有的路径,查清楚之前不动
可恢复优先 能进废纸篓就不直接删
状态不碰 登录态、聊天记录、钥匙串,永远不动
先退出再动手 app 还在跑就先关掉,避免边删边重建
sudo 交给你 系统级删除只生成命令,你自己跑

安装

git clone https://github.com/orange2ai/disk-raccoon ~/.claude/skills/disk-raccoon

支持 SKILL.md 协议的 agent 都能用(Claude Code、Cola 等)。

用法

跟 agent 说:

实战记录

缓存 释放
~/.cache/uv(Python 包缓存) ~56G
Chrome Service Worker + ChromeDebug ~51G
飞书 LarkShell/aha 文档缓存 ~38G
合计 135+ GB

参与

缓存地图是活的,我们每清一次新缓存就补一条进去。如果你也踩过新的坑,欢迎提 issue 或 PR,把更多缓存加进来,让这只浣熊越来越懂你的硬盘 🦝

License

MIT

disk-raccoon (orange2ai/disk-raccoon)

Searching Database…

What is this?Disk Raccoon (硬盘小浣熊) is an AI skill that helps your coding agent find macOS cache folders eating up disk space, figure out which ones are safe to delete, and clean them for you.
Is this a standalone cleaning app, or something I add to my AI agent?It is a SKILL.md-based skill, not a standalone app. You load it into an AI agent that supports the SKILL.md protocol, such as Claude Code, and the agent does the analysis and cleanup.
How do I use it?Install it as a skill in a SKILL.md-compatible agent, then just ask in plain language like 'help me check my cache' or 'disk is full, what can I clean', and the agent walks the cache map and reports back.
Which AI agents can run this skill?Any agent that supports the SKILL.md protocol; the README names Claude Code and Cola as examples.
Does it delete files automatically without asking?No - the first pass is read-only and only produces a report. Paths not yet classified in the cache map are left untouched until verified, and anything needing sudo only gets a command printed for you to run yourself.
How does the cache map work?It is a living reference file (references/cache-map.md) where every known cache path is tagged with a safety level: safe to delete, delete with care, or never touch. Unknown paths get investigated before any action is taken.
Why is it called a raccoon?Raccoons are famous for washing food before eating it - the project applies that same check-before-you-touch caution to cache cleanup instead of blindly deleting whatever is largest.
What has it actually cleaned so far?In its first real run it freed over 135GB: about 56GB from the uv Python package cache, 51GB from Chrome's Service Worker and debug cache, and 38GB from Feishu's LarkShell document cache.
Creator
orange2ai
Size
<0.1 MB
Stars
203
Updated
CommitAug 31
Age
1mo
Commits · 90 Days