InertiaRSS Track and read blogs, news, and tech you care about
Read Original Open in InertiaRSS

Recommended Feeds

博客园 - 司徒正美
V
V2EX
T
Tailwind CSS Blog
有赞技术团队
有赞技术团队
aimingoo的专栏
aimingoo的专栏
Apple Machine Learning Research
Apple Machine Learning Research
IT之家
IT之家
Blog — PlanetScale
Blog — PlanetScale
A
About on SuperTechFans
月光博客
月光博客
T
The Blog of Author Tim Ferriss
宝玉的分享
宝玉的分享
Martin Fowler
Martin Fowler
博客园 - 聂微东
The GitHub Blog
The GitHub Blog
V
Visual Studio Blog
WordPress大学
WordPress大学
酷 壳 – CoolShell
酷 壳 – CoolShell
Engineering at Meta
Engineering at Meta
GbyAI
GbyAI

阮一峰的网络日志

科技爱好者周刊(第 396 期):互联网通信的替代方案 科技爱好者周刊(第 396 期):互联网通信的替代方案 - 阮一峰的网络日志 科技爱好者周刊(第 395 期):软件开发的第三种方式 科技爱好者周刊(第 395 期):软件开发的第三种方式 - 阮一峰的网络日志 科技爱好者周刊(第 393 期):脑腐状态 科技爱好者周刊(第 392 期):axios 投毒与好莱坞式骗术 科技爱好者周刊(第 391 期):AI 的贫富分化 科技爱好者周刊(第 390 期):没有语料,大模型就是智障 套壳中国大模型撑起500亿美元估值?扒一扒 Cursor 的"套壳"疑云 科技爱好者周刊(第 389 期):未来如何招聘程序员 科技爱好者周刊(第 388 期):测试是新的护城河 零安装的"云养虾":ArkClaw 使用指南 科技爱好者周刊(第 387 期):你是领先的 科技爱好者周刊(第 386 期):当外卖员接入 AI 科技爱好者周刊(第 385 期):马斯克害怕中国车企吗? 智谱旗舰 GLM-5 实测:对比 Opus 4.6 和 GPT-5.3-Codex 科技爱好者周刊(第 384 期):为什么软件股下跌 科技爱好者周刊(第 383 期):你是第几级 AI 编程 Kimi 的一体化,Manus 的分层 科技爱好者周刊(第 382 期):独立软件的黄昏 AI native Workspace 也许是智能体的下一阶段 科技爱好者周刊(第 381 期):中国 AI 大模型领导者在想什么 科技爱好者周刊(第 380 期):为什么人们拥抱"不对称收益" 科技爱好者周刊(第 379 期):《硅谷钢铁侠》摘录 我如何用 AI 处理历史遗留代码:MiniMax M2.1 升级体验 科技爱好者周刊(第 378 期):预测是新的互联网热点 科技爱好者周刊(第 377 期):14万美元的贫困线 科技爱好者周刊(第 376 期):太空数据中心的争议 科技爱好者周刊(第 375 期):一扇门的 Bug 终于有人做了 Subagent,TRAE 国内版 SOLO 模式来了 科技爱好者周刊(第 374 期):6GHz 的问题 VS Code 使用国产大模型 MiniMax M2 教程 科技爱好者周刊(第 373 期):数据模型是新产品的核心 国产大模型接入 Claude Code 教程:以 Doubao-Seed-Code 为例 科技爱好者周刊(第 372 期):软件界面如何设计 大模型比拼:MiniMax M2 vs GLM 4.6 vs Claude Sonnet 4.5 科技爱好者周刊(第 371 期):一个乐观主义者的专访 科技爱好者周刊(第 370 期):正确的代码高亮 错误处理:异常好于状态码 科技爱好者周刊(第 369 期):Tim 与罗永浩的对谈 科技爱好者周刊(第 368 期):不要这样管理软件团队 一天之内,智谱和 Anthropic 都发了最强编程模型 科技爱好者周刊(第 367 期):Nano Banana 的几个妙用 科技爱好者周刊(第 366 期):旧金山疯狂的 AI 广告 科技爱好者周刊(第 365 期):流量变现正在崩塌 科技爱好者周刊(第 364 期):最难还原的魔方 科技爱好者周刊(第 363 期):最好懂的神经网络解释 科技爱好者周刊(第 362 期):GitHub 工程师谈系统设计 科技爱好者周刊(第 361 期):暗网 Tor 安全吗? 科技爱好者周刊(第 360 期):Dan Wang 的新书
Byte Family Bucket Seed 2.0 + TRAE Mastering Skill
阮一峰 · 2026-02-14 · via 阮一峰的网络日志

I. Introduction

Among domestic large models, ByteDance is an outlier.

Unlike other large models that make a big fuss and fight for attention, it is more low-key and less noticeable.

However, it does the most, with independent brands for large models, Agents, development tools, and cloud services, all blossoming everywhere, covering all bases and advancing at high speed.

Seed is ByteDance's large model team, with several product lines under it, and the recently hyped video model Seedance 2.0 is their product.

Today, I will write a Skill tutorial using ByteDance's full suite ---- the newly released Seed 2.0 model and development tool TRAE.

You will see that they work together as powerful yet simple and user-friendly, and (for personal users) are also free This is also why I wrote this, to let everyone know about this solution.

It only takes ten minutes to read this tutorial, and you will also understand what Skill is, how to use it, and why you must use it.

II. Introduction to Seed 2.0

First, let's introduce Seed 2.0, which is the base model of the Seed family.

A "base model" (foundation model) refers to a general-purpose large model, which can be used to build various downstream models. The two main features are scale and strong generalization ability, making it easier to build other models.

The well-known Doubao is based on the Seed model and is also known as the "Doubao Large Model". This time, Seed 2.0 includes three general-purpose models: Pro, Lite, and Mini, as well as the Seed 2.0 Code model specifically designed for developers.

As it must support various uses, Seed 2.0's versatility is particularly prominent, stronger than previous versions.

1. Supports multimodal, capable of processing various types of data: text, charts, visual space, motion, video, and more.

2. Possesses various Agent capabilities, facilitating integration with enterprise tools: search, function calls, tool calls, multi-round instructions, context management, etc.

3. Has reasoning and coding capabilities.

Due to the last point, we can use it for programming, especially for generating front-end code. Combined with the AI programming tool TRAE released by ByteDance, the effect is excellent, especially convenient for full-stack development, and it's free for individual users.

III. Preparations for TRAE

Download and install TRAE, it has two modes, which can be switched in the upper left corner: IDE mode and SOLO mode.

Select IDE, SOLO is an AI task orchestrator, unless multiple tasks are run together, it's not needed.

Then, press the shortcut key Ctrl + U (or Command + U) to bring up a dialog box for conversation with AI.

We are building a web application, so select @Builder development mode in the upper left corner. Choose Seed-2.0-Code in the lower right corner.

It can be seen that TRAE's built-in domestic open-source programming model is very comprehensive and all free to use.

Preparation is almost done.

Four, programming and testing

I chose a somewhat challenging task to generate with Seed 2.0.

ASCII art is graphics drawn using characters, such as the image below.

I plan to create a web application where users can input ASCII graphics on the webpage, which will automatically convert them into hand-drawn style graphics in the Excalidraw style.

Prompt as follows:

"Create a web app that can convert ASCII graphics into Excalidraw-style images and provide a download option."

The model begins to think, breaking down this task into four steps.

V. Generate Results

Once the code generation for Seed 2.0 is complete, TRAE will start a local service at localhost:8080, while also opening the preview window.

The generated result is quite interesting, the upper ASCII input box provides four examples: Box, Tree, Flowchart, Smiley. Below is the appearance of Tree.

Then comes the control panel for Excalidraw parameters: line width, roughness, curvature, font size.

Click the Convert (Convert) button to get a hand-drawn style line drawing immediately.

The entire page looks like this.

VI. Skill Introduction

The design of this page doesn't feel very aesthetic and could be improved. I plan to add a dedicated front-end design skill to Seed 2.0 to create more aesthetically pleasing pages.

A "Skill" is a prompt specifically designed for a particular purpose, used to inject context.

Sometimes, prompts are very long, and typing them every time is cumbersome. We can extract the frequently used parts and save them in a file for easy reuse. Such extracted prompts are often detailed descriptions of how to complete a task, so they are called "skill files."

In terms of format, it is a Markdown text file with a YAML header containing the name field and description field.

The name field is the name of the Skill, which can be used to invoke the skill by this name; the description field is a brief description of the skill, and the model uses this description to determine when to automatically invoke the skill.

Some skills are quite complex, in addition to the description file, there are dedicated script files, resource files, template files, and so on, similar to a codebase.

Among these files, SKILL.md is the entry file, and the model uses its description to understand when and where to invoke the other files.

This library can be shared online, allowing others to use it. If you find that the AI model needs a certain skill when processing tasks, you can look for Skills written by others and load them into the model.

Section 7: Front-end Design Skills

Below, I use what Anthropic Inc. has shared.Front-end design skillsRedesign the previous page. It only consists of a single Markdown file, which can be downloaded.

Open the "Settings/Rules and Skills" page for TRAE.

Click the "+ Create" button in the Skills section to open the skill creation window.

You can fill in the SKill content in this window or upload an existing Skill file. I choose to upload, and after completion, you can see that the frontend-design skill is already in the list.

Then, I used the following prompt to invoke this skill to rebuild the page.

"Using the frontend-design skill, refactor this page to make it more beautiful and user-friendly, with a more professional feel."

Below is the text description and reconstruction result provided by the model.

The page indeed feels more upscale!

8. Vercel Deploy Skill

Finally, let's look at another example of a skill.

After code generation, it runs on the local machine. Can it be published online and shared with more people?

The answer is that as long as you use Vercel's deploy skill, you can publish the generated results to Vercel's machines with a single command.

In the GitHub repository of Vercel's official skills, download the zip file for the Vercel-deploy skill.

Then, drag this zip file into the skill window of TRAE, and it will load automatically.

"Publish the generated website to Vercel."

The model will execute the vercel-deploy skill, publish the website to Vercel, and finally provide two links: one is the preview link, and the other is the link to publish to your personal account.

Everyone can access nowThis linkLet's see the actual effect of the website.

IX. Summary

If you've read this far, you should agree with my point. Seed 2.0's programming capabilities are quite good, and when paired with its own programming tool TRAE, it's both user-friendly and free.

Skill is a powerful capability extension mechanism that makes the model capable of anything, so it's essential to learn how to use it.

(Complete)