前言
以前总有一些想做但是没做的小工具，想法攒了一堆，真到动手的时候又开始嫌麻烦：要搭环境、要写界面、要处理各种系统差异，最后往往就放下不了了之。然后最近有时间的时候，就在家让AI来实现，我只负责出需求和核对检测，然后使用了一段时间之后，我觉得这些插件功能不错，于是想分享一下，但是这次不想手写博客了，顺便让AI根据我的风格写出了这篇博客，。
一、先说下这三款插件是做什么的
三个插件都放在同一个仓库里，用的是 uTools 的插件规范，装完之后在主输入框里敲一个词就能打开。
- 电脑配置体检：读取本机的 CPU、内存、硬盘、显卡、主板和系统信息，给出性能评分、与主流配置的对比、硬件天梯榜的排名定位，以及升级建议；
- 开发配置速查：把散落在各处的开发配置文件一次性找齐，包含 Git/SSH、npm/yarn/pnpm/nvm、Maven/Gradle、hosts，以及 Codex、Qoder、WorkBuddy、Trae、VS Code 这些 AI 工具和编辑器的配置；
- 电脑空间管家：磁盘占用分析、大文件排行、重复文件识别、视频与图片的压缩瘦身、垃圾清理、软件搬家、内存与进程查看、游戏目录分析、文件夹整理。
整体思路其实很朴素：把"我自己会用到"这件事放在第一位，界面尽量一眼看懂，涉及删除这类操作就多设一道关卡。 下面分别说一下每个插件的功能和用法。
二、为什么要做这三个插件
这三个插件都不是"想到了才做"的，每一个背后都对应一件真实的麻烦事。
1. 我的主力机想升级，但不知道先换哪个
我这台机器是 i5-8400 + 8G 单通道内存 + 机械硬盘 + GTX 1050 Ti，用了好几年了，一直想升级，但每次打开购物网站又犹豫：先换 CPU 还是先加内存？换显卡值不值？
网上的跑分软件也用过一些，要么是给整机打一个看不懂的分数，要么是硬件信息列了一堆但看不出好坏，用完之后问题还在。所以我想要的东西其实很简单：告诉我这台机器处在什么水平、最该换的是什么。
2. 每次换电脑、装新工具，配置都找不到
配置文件的麻烦在于它太散了。.gitconfig 在家目录，.npmrc 在家目录（有时候也在项目里），Maven 在 .m2 下面，hosts 在系统目录，前端还有 yarn、pnpm、nvm 各自的一份；
这两年 AI 工具多了之后更夸张，Codex、Qoder、WorkBuddy、Trae、Cursor 每家都有自己的配置目录和规则文件。
所以我想要的也很简单：有个地方能一次看全，能搜、能定位，需要改的时候能直接改。
3. C 盘又红了，但清理软件不太敢用
清理类的软件我用过一些，最不放心的就是它删东西之前不告诉你要删什么，点一下"一键清理"，几十 G 就没了，也不知道删的是什么。而我真正想清理的往往是"那个 10G 的录屏文件"这种具体的东西，不是系统猜出来的垃圾。
所以空间管家的定位从一开始就很明确：先让我看清有什么可以清，再让我一条条确认。
三、我是怎么用 AI 把插件写出来的
这部分应该是这篇博客里最有价值的部分了。三个插件我是用 WorkBuddy 这类能直接读写本地代码的AI 编程助手做的，整个过程和我以前自己写代码的路子差别挺大，我大概分成这么几步。
1. 先讲清楚"我要什么"，而不是先讲怎么写
我以前用 AI 就是丢一句"帮我写个功能"，然后来回改，改到最后它也不知道我到底想要什么。 这次我换了个方式，一开始就跟它把三件事讲清楚：
- 场景：什么时候会用到（比如"C 盘满了，我想先看看哪个目录最占空间"）；
- 界面：我打开插件第一眼要看到什么（比如"首页不要一进来就扫描，先给我几个任务入口"）；
- 标准：什么叫做好了（比如"系统盘用了 80% 以上要说出来""机械硬盘要明确告诉我是瓶颈"）。
这三件事讲清楚之后，AI 给出的东西基本就是我想要的样子，剩下的是细节调优。 我后来发现，插件页面的布局、配色、交互这些它做得比我快，我不用在这一块花时间了。
2. 让它先去查官方文档，别自己造轮子
uTools 已经内置了不少能力，比如 FFmpeg 7.1 做音视频处理、Sharp 做图像处理，这些通过插件 API 直接调用就行，不需要自己引入依赖。
所以我给它定了条规矩：写功能之前先去查 uTools 的 API 文档有没有现成的。
后来我还让它把官方文档整个抓成离线副本放进仓库，这样后面每次做新功能，它都能先查一遍再动手。空间管家里的视频压缩就是直接用 utools.runFFmpeg 做的，连压缩进度都是官方接口自带的，省了很多事。
3. 让它顺手把自测脚本和截图脚本也写了
这是我觉得最值的一步。
AI 写代码很快，但如果你只是让它一直加功能，不出一周自己就不敢改了——因为你不知道哪里坏了。 所以我让它每写一块功能，就顺手写一个能跑的验证脚本。现在仓库里是这样分层的：
- 数据层：纯 Node 脚本，验证数据表是否完整、搜索口径对不对；
- 桥接层：把 uTools 的接口打桩，验证文件读写和安全约束；
- 界面层：用无头浏览器把真实页面跑起来，模拟真实点击，再看状态有没有变。
界面这一层我觉得特别有用，因为很多问题静态看代码是看不出来的。脚本在后台跑一遍，报告里会写清楚哪一项没过，比我自己一个个点快多了。
商店截图也是同理。我让 AI 把截图也做成脚本：先用无头浏览器打开页面、切到指定的标签页、把窗口高度开大、截图，最后再把底部多余的空白裁掉。现在要更新素材，改一行配置重新跑一次就行，十几张图自动生成，不用我一张张去截。
4. 跑真实数据，口径由我来定
AI 很会算，但"什么才算好"这件事它不会替我做决定，这部分我一直是自己拍板的。
举个实际的例子：配置体检的评分，我定的口径是 100 分对应 2026 年的高端主流配置，综合分按处理器 30%、内存 20%、存储 20%、显卡 20%、系统 10% 加权。 这个权重是我按自己的使用感受定的，不是它算出来的。
再比如天梯榜，我坚持只用消费级的数据——服务器 CPU 不放进榜单，否则家用机的位置会被压得过低；拿不到公开评测数据的型号，界面上会标一个 ≈，说明这是推算值，不能当成精确成绩去引用。
空间管家那边也是一样。删除这个能力它做得很顺手，但我给它定的规则是： 只允许删除本次会话里出现过的文件、默认只能删到回收站、彻底删除默认关闭、执行前必须逐条勾选确认。这些不是技术问题，是"要不要为用户的误操作负责"的问题，这种取舍我不放心交给 AI。
5. 按版本迭代，需求我来提
三个插件都不是一次写完的，基本是"自己用几天，发现不顺手就改"。举几个我自己提的需求：
- 空间管家打开就自动扫描太吵了，改成打开不扫描，首屏先给任务入口（v1.2.0 → v1.3.0）；
- 压缩完之后产物不知道跑哪去了，改成写回源文件所在目录，并按分辨率命名，比如 1.mp4 → 1-720p.mp4（v1.2.0）；
- 大文件页只能一个个勾太累，加了全选本页、勾选当前筛选全部，还有跨页累积（v1.3.0）；
- 配置体检做完之后少一个"横向比"的参考，加了硬件天梯榜，能看自己的机器排在第几（v1.1.0）。
提需求这个过程反而变简单了，因为不用再组织语言去描述实现，直接把感受说出来就行。
6. 一点体会
用了这段时间，我最大的感受是：AI 很强，但它没有主观。
强的地方在于，写界面、处理各种系统差异、写脚本这些我以前嫌烦的活，它做得比我快； 不强的地方在于，它不知道"这个功能做出来会不会让人误删文件"，也不知道"这个评分口径对用户是不是公平"。这些还是得人来定。
所以我现在的分工是：它写代码，我定规则和验收。它写完我不做逐行 review，但我会自己跑一遍、用真实数据看结果对不对，发现不对就说清楚哪里不对，让它改。 这样留一步参与感也挺好的，至少我知道自己的东西是怎么实现的。
四、电脑配置体检
这个插件的核心就是把系统信息采集上来之后，翻译成人能看懂的话。
打开之后第一屏就是结论，不绕弯子：
我随便拿自己的旧机器举例：综合评分 35 分，定位在"入门办公"档位，每张评分卡除了分数之外，还会给一句解释，比如存储那一项直接写着"系统跑在机械硬盘上是整机最严重的瓶颈"。
对比页是把本机放到几个档位里横向比：
从入门办公到旗舰一共五档，每一档给出参考配置和分数差距，这样一眼就能看出来"再加多少钱能到什么水平，值不值"。
天梯榜是我个人最喜欢的一块，能看出来自己的机器在市场上是什么位置：
一共有五个榜单：CPU 多核、CPU 单核、显卡、内存、硬盘，加起来两百多条数据。 本机会带着"本机"的标记直接插进榜单里，卡片上会写明"第 57 / 63 名， 超过 10% 的机型"。
五、开发配置速查
这个插件解决的是"配置文件在哪"和"顺手改一下"这两件事。
一共整理了 9 个分类 55 条配置，左边是分类树，右边是详情。顶部可以搜索，也可以用"仅存在""仅敏感"这类筛选条件把列表收窄。每条后面有个小圆点表示这台机器上有没有这个文件，敏感的条目会带一把锁。
我这两年配置里放的东西越来越多，密钥和凭据也不少，所以这个插件在安全上做了几件事：
- 像 SSH 私钥、Git 凭据、各家 AI 工具的授权信息这类文件，会被自动识别出来， 预览默认打码，要看的话得手动点一下；
- 敏感文件不提供内置编辑，只保留"用系统程序打开"的引导；
- 可编辑的配置文件，每次保存之前自动备份一份，保留最近 20 份，随时可以还原；
- 如果文件在插件读取之后被别的程序改过，保存会中止并提示，不会覆盖别人的改动。
平时比较常用的其实是"复制路径"：知道配置在哪之后，剩下的用编辑器去改更顺手。
六、电脑空间管家
这个插件是三个里面最花时间的，功能也最多，目前已经有十个页面。
首屏我没有做成扫描结果，而是做成了任务入口，因为我发现大多数时候我并不想扫全盘：
压缩、垃圾清理、软件搬家、内存与进程、游戏目录、文件夹整理都能直接开始，入口上还会带上实时的角标，比如"垃圾清理 已扫到 20.0 GB""软件搬家 可迁移 10 个"。 只有"磁盘占用分析"需要真的扫描一遍。
不扫描的时候首页是这样一个空状态，不会让人以为插件卡住了：
压缩是我用得最多的功能，可以直接挑单个文件、多个文件，也可以挑整个文件夹批量压：
方案表里会列出每个文件的分辨率、编码参数、预计压缩后的体积和节省比例， 最上面还能看到合计能省多少。几个我比较在意的点：
- 压缩结果写回源文件所在目录，并按分辨率命名，比如 1.mp4 → 1-720p.mp4，
图片 → 封面-1920x1080.webp，不会再出现"压完了找不到文件"；
- 永远输出到新文件，不会覆盖原文件，原文件保留还是清理由我自己决定；
- 这里的压缩都是视觉无损的思路，不是数学意义上的无损。 只要改分辨率或者重新编码就一定会丢信息，界面上也把这句话写在明面上了。
清理这块是我花心思最多的地方，因为删文件这事最容易出事：
删除之前一定会弹这么一张清单：每个文件的完整路径、体积、修改时间都列出来，合计能释放多少也写清楚，要勾上一句"我已逐条核对上述文件"才能点删除。 默认是删到系统回收站，想彻底删除还得单独开一个开关。被安全策略拦下来的文件也会列出来，说明为什么不能删，而不是悄悄少一项。
其他几个功能简单说一下：
- 垃圾清理：临时文件、浏览器缓存、Windows 更新残留、回收站这些，扫完给清单和体积；
- 软件搬家：把装在系统盘的程序迁到别的盘，原位置留一个链接，程序照旧从原来的路径启动， 不需要改注册表，中途失败会自动回滚，源目录始终可用；
- 内存与进程：按占内存排序看进程，可以直接结束占内存的大户。 这里要说句实话，内存是不能被"释放"的，只能通过结束进程来回收；
- 游戏目录：识别 Steam 和 Epic 的游戏库并统计占用，只做分析；
- 文件夹整理：把一个乱目录按类型或者时间归档到子文件夹里，执行前先给预览。
顺手说下性能，扫描这块我实测过：在自己用户目录下，56 万个文件、6.6 万个目录、45.2 GB， 大概 40.8 秒扫完，扫描过程中可以随时停。
七、怎么安装和使用
需要先装 uTools 和 uTools 开发者工具，两个都在官网可以下到。
- 打开 uTools 开发者工具，点"新建项目"，项目路径选择插件目录（就是包含 plugin.json 的那一层）；
- 点击"接入开发"；
- 回到 uTools 主输入框，输入插件对应的指令就能打开。
三个插件的常用指令如下：
另外，这三个插件都注册了 AI 工具接口，也就是说可以直接在 AI 里问"我这台电脑什么水平""我的 npm 配置在哪"， 不需要手动打开插件。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Foreword
There have always been little tools to do, but not to do, and ideas have accumulated, and when it comes to doing it, it starts to be difficult: to set up the environment, to write interfaces, to deal with system differences, and often to leave them behind. And recently, when I had time at home, I had to take out demand and check, and after a while, I thought the plugs were working well, and I wanted to share them, but this time I didn't want to write a blog by hand, and I had it written by my style.
First, tell me what these three plugs do.
All three plugins are in the same warehouse, using the uTools plugin code, which can be opened with a word in the main input box.
- Computer configuration check-ups: reading CPU, memory, hard disk, GPU, masterboard and system information on the machine, giving performance ratings, comparison with the mainstream configuration, ranking of the hardware ladder, and recommendations for upgrading;
- Development of configuration swaths: one-time search of development profiles scattered throughout the sites, including Git/SSH, npm/yarn/pnpm/nvm, Maven/Gradle, hosts, and the configuration of Codex, Qoder, WorldBudy, Trae, VS Code, these AI tools and editors;
- Computer space controllers: disk occupancy analysis, large file rows, duplicate document identification, compact thin bodying of videos and pictures, garbage removal, software removal, memory and process viewing, game catalogue analysis, folder preparation.
The whole idea is simple: putting "I'll use it myself" first, the interface is as simple as possible, and there's an additional level involved in deleting this type of operation. The functions and usage of each plugin are described separately below.
II. WHY THE THREE Plugins
None of these three plugins are "for thought" and each one of them is a real problem.
1. My main force wants to upgrade, but I don't know which one to replace.
My machine is i5-8400 + 8G single channel memory + mechanical hard drive + GTX 1050 Ti, has been trying to upgrade for years, but every time I open a shopping site, I hesitate: change CPU or add memory? Is GPU worth anything?
Online rundowns have also been used, either to give the whole machine an intelligible score, or to list a pile of hardware information that does not appear to be good or bad, and the problem remains after that. So what I want is really simple: tell me what level this machine is and what it's worth.
2. Cannot find configurations for every computer change and installation
The problem with the configuration file is it's too loose....gitconfig home directory,.npmrc home directory (sometimes also in the project), Maven below.m2, hosts in the system directory, front end, one copy each;
Over the last two years, the number of AI tools has increased, and Codex, Qoder, WorkBuddy, Trae, Cursor each has its own configuration catalogue and rule files.
So what I want is simple: there's a place to look at at at once, to search, to locate, to change when it needs to.
The C drive is red, but the cleaning software isn't very good.
I've used some cleaning software, and the most unsure thing about it is that it doesn't tell you what to delete before it deletes, and it clicks "one-click cleans," and dozens of Gs are gone, and it doesn't know what to delete. And what I'm really trying to clean up is that "the 10-G video file" specific stuff that's not system-gum.
So the space housekeeper's location was clear from the beginning: first let me see what's clear, then let me confirm.
Three, how did I write the plugin with AI?
This part should be the most valuable part of this blog. The three plugins I did with an AI programming assistant like WorkBuddy, who can read and write local code directly, the whole process is quite different from the way I used to write my own code, and I'm about to split it into these steps.
I want to make it clear what I want, not what I want first. Write
I used to use AI to drop a "write me a function" and then change it back and forth, and eventually it didn't know what I wanted. This time I changed the way I started by telling it three things:
- Scene: When will it be used (e.g., "C's full, I want to see which directory is the most occupied space");
- Interface: I opened the plugin to see at first sight (e.g. "Don't scan the front page and give me a few mission entrances");
- Standard: what's good?
After these three things were made clear, AI gave basically what I wanted, and the rest was the finer details. I later found that the layout, coloring and interaction of the plugin pages were faster than I did, and I don't need to spend time in this area.
Let him check the official files first, not make his own wheels.
uTools has built in a number of capabilities, such as FFmpeg 7.1 for audio and video processing and Sharp for image processing, which can be called directly through Plugin API without having to introduce dependency.
So I set a rule for it: check if the uTools API files are available before writing.
Later, I made it catch the entire official document as a copy of offline and put it in the warehouse so that every time it had a new feature, it could check it out and do it again. The video compression in the space pipe house was made directly from utools.runFFmpeg, even with the official interface, saving much.
Let it write the script and the screenshot.
This is the step I think is most valuable.
AI wrote the code quickly, but if you just let it stay functional, you won't change it in a week -- because you don't know where it's broken. So I made it write a run-through script for every single feature. Now there's a layer in the warehouse:
- Data layer: pure Node script to verify whether the data sheets are complete and search calibrated;
- Bridge layer: pillar the uTools interface to verify document reading and writing and security constraints;
- Interface level: run the real page with a headless browser, simulate real hits and see if the situation has changed.
This level of interface I find particularly useful, because a lot of questions are static looking at code is not visible. Scripts run backstage, and the report will make it clear which one is not, much faster than myself.
The shop's cut-off is the same thing. I asked AI to make the screenshot into a script: open the page with a headless browser, cut to the specified tab page, expand the height of the window, cut the screenshot, and then cut the extra blanks at the bottom. Now we're going to update asset and rerun it in one line.
I'll fix the calibration.
AI's good at counting, but "what's good" is not what it's gonna decide for me, and I've been making this part myself.
To give a practical example: I set a calibration of 100 points for the high-end mainstream configuration in 2026, divided by 30% of processors, 20% of memory, 20% of storage, 20% of GPU, 10% of system weight. This weight is what I feel for my own use, not what it counts for.
And for example, I insist that only consumer-grade data -- servers, CPUs don't put on the list, or the position of the home machine will be overpressed; without a public rating model, the interface will be marked with an ellipse, indicating that this is a extrapolation and cannot be quoted as an exact accomplishment.
So is the space housekeeper. It's easy to delete this capability, but I've given it rules that allow only those documents that have appeared in this session to be deleted, that the default can only be deleted to the wastebin, that the default closure is deleted completely, and that the article-by-article confirmation is required before execution. It's not a technical question, it's a question of whether you're responsible for the user's mishandling. I don't trust AI.
5. I'll take care of the needs in the version.
None of the three plugins were finished at once, basically "for a few days, you can change it if you don't like it." A few of my own needs:
- Auto-scanning when the space housekeeper opens is too loud to open without scanning, with the first screen given to the mission entrance (v1.2.0 → v1.3.0);
- The product went nowhere after the compression was completed, was changed to the directory where the source document was written and named after resolution, e.g. 1.mp4 → 1-720p.mp4 (v1.2.0);
- Large pages can only be filled with one sheet, with a full selection of the pages, a checkout of the current selection, and a cross-page accumulation (v1.3.0);
- A "horizontal comparison" reference is missing after the medical check-up has been completed, with the addition of a list of hardware ladders, allowing for the viewing of one's own machine at the top (v1.1.0).
The process of raising demand has become simpler, as there is no need to organize language to describe realization and simply to say what it feels.
A little appreciation.
My biggest feeling in this time is that AI is strong, but it's not subjective.
The strong part is that I'm tired of writing interfaces, dealing with system differences and writing scripts, and it's doing it faster than I used to do; the weak part is that it doesn't know whether "this function is going to make you miss the file" or whether "this rating is fair to users." It's still up to you.
So my division now is: it writes code, I make rules and acceptance check. I don't do line-by-line reviews, but I'll run through it myself and see the results with real data, right? It's a good way to stay involved, at least I know how my stuff works.
Computer Configuration Medical Examination
The core of this plugin is to get the system information and translate it into words that one can understand.
When you open the first screen, it's a conclusion.
I take the example of my old machine: a comprehensive score of 35, located in the "Into-house" slot, and each score card, in addition to a score, explains, for example, the one that's stored and says, "The system runs on a mechanical hard drive is the biggest bottleneck of the machine."
The contrast page is a horizontal comparison of the machine in several slots:
From entry to flagship, there are five tranches, each showing a gap between reference configurations and scores, so that it can be seen at a glance "how much more can be added to the level, which is worthless."
The ladder is one of my personal favorites, showing where my machine is on the market:
There are five lists: CPU multi-nuclei, CPU mononuclei, GPU, memory, hard drives, which add up to over 200 data. This opportunity is inserted directly into the list with the tag "The Machine" and the card says, "57rd 63rd place, more than 10% of the machine."
V. Development of a mapping exercise
This plugin solves the two things: "Where's the configuration file?" and "replace it."
Nine categories were sorted, 55 configurations on the left were sorted trees and details on the right. The top can be searched, or the list can be narrowed down with a filter condition such as "just exist" "just sensitive". There's a small dot at the back of each of these, which means there's no file on this machine, and sensitive entries have a lock.
I've put more and more in my configuration over the last two years, and I have a lot of keys and documents, so this plugin has done a few things in safety:
- Files such as SSH private keys, Git certificates, and authorized information from various AI tools are automatically identified, previewing default numbers and showing them manually;
- Sensitive documents do not provide built-in editing and retain the "open with a system program" guide;
- Editable profiles, which are automatically backed up before each saver, keeping the most recent 20 and ready to be restored;
- If the file is modified by another program after the plugin is read, the saving will be suspended and prompted and no other changes will be covered.
What's more common is "replicate paths": after knowing where the configuration is, the rest of the editor is going to do better.
Computer space housekeepers
This plugin is the most time-consuming of the three and most functional, and currently has 10 pages.
I didn't do the scan on the front screen, but I did the mission entrance, because I found that most of the time I didn't want to clean it up:
Compression, garbage removal, software movers, memory and process, game catalogues, folders, and real-time corner signs on the entrance, such as "waste cleanup to 20.0 GB" software movers, can move 10." Only Disk Occupancy Analysis needs a real scan.
The first page is empty without scanning and does not seem to be stuck:
Compression is the most useful function I can use to select single files, multiple files or a whole folder mass pressure:
The table shows resolution, coding parameters for each document, the volume expected to be compressed and the proportion of savings that can be saved at the top. A few things I care more about:
- Compression returns the directory where the source file is located and is named after resolution, e.g. 1.mp4 → 1-720p.mp4,
Photo: Cover 1920x1080.webp, "No files found after crushing"
- Always output to a new document, which does not overwhelm the original document, which is left to me or to clean up;
- All the compression here is visual, not mathematical. If resolution is changed or recoded, the message will be lost, and it will appear on the interface.
Cleaning up this piece is the best thing I've ever thought about, because it's easier to delete documents:
This list will always be played before it is deleted: the complete path, volume, modification time of each document is shown, the total amount of releases is clearly written, and the last sentence, "I have checked the above-mentioned document article by article" can only be deleted. The default is to delete to the system wastebin, and a single switch is required to remove it completely. Documents that have been stopped by security tactics are also listed, explaining why they cannot be deleted, rather than missing one quietly.
The other functions are brief:
- Waste disposal: temporary file, browser cache, Windows Update Residues, Wastebins, etc., to be completed with inventory and volume;
- Software movers: move the application installed on the system disk to another disk, leave a link in its original location, start the program from the original path as usual, do not need to change the registration form, the stopover failure automatically rolls back and the source directory remains available;
- Memory and process: the process is ranked according to the proportion of memory, which can directly end the large households that account for it. To be honest, memory cannot be "released" and can only be recovered by ending the process;
- Catalogue of games: identify the game libraries of Steam and Epic with statistical occupancy, only for analysis;
- Folder preparation: Archive a random directory into a subfolder by type or time, previewing it before executing it.
By the way, I've measured the performance: 560,000 files, 66,000 directories, 45.2 GB, about 40.8 seconds to clear, and can stop at any time during the scan.
How to install and use
You need to install uTools and uTools Developer tools first, both of which are available online.
- Open the uTools Developer tool, click "New Project", and the project path selection plugin directory (the floor containing plugin.json);
- Click "access development";
- Back to the uTools main input box, enter the instructions corresponding to the plugin.
Common instructions for three plugins are as follows:
In addition, all three plugins are registered for the AI tool interface, which means they can ask "What level of my computer" directly in AI where my npm configuration is" without having to open the plugin manually.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.