WorkBuddy Case Study: Two Browser Games, One Published
A Juejin author used Tencent’s WorkBuddy to build two browser games from natural-language requests, producing single-file HTML applications with Canvas-based logic, graphics and sound. According to the author’s local session and credit records, the deep-sea fishing project took 43 conversations and 301.46 credits, while the forest combat game took 16 conversations and 107.59 credits. The fishing game was published through WorkBuddy’s sharing feature; the forest game remained local. The author reported that gameplay and visual changes required repeated instructions, automated submission of a webpage form failed and was completed manually, and mobile landscape support required rotating the display with CSS and converting touch coordinates.
The report provides session-backed measurements of a practical AI-assisted browser-game workflow, including credit costs, generated files, publication steps and documented failures and workarounds. Its findings may help practitioners understand where this single-user workflow required iteration, but they are not an independent evaluation of WorkBuddy.
导读：上一篇 WorkBuddy 讲了它是什么、积分怎么领。这篇只讲一件事：用它做网页小游戏。我在 10-29 到 11-02 之间用它做了两个游戏，一个深海钓鱼、一个森林砍怪，WorkBuddy 本地的会话记录和积分表都还在。我把每一次对话花了多少积分、调了多少次工具都拉了出来，整理成这篇：第一版要多少积分、钱花在哪种需求上、怎么说话最省，以及一个所有网页游戏都会碰到的坑：发布后在手机上横不过来。
一、先说清楚：用 WorkBuddy 做小游戏是怎么回事
WorkBuddy 是腾讯出的桌面 AI 助手，能在你电脑上读写文件、跑命令。做小游戏时，你用中文描述想要什么，它写出一个单文件 HTML（游戏逻辑、画面、音效都在这一个文件里，用浏览器 Canvas 画），自己打开测一遍，再用自带的分享功能发布成一个 workbuddy.link/p/<作品 ID> 链接，别人点开就能玩，不用登录。
整个过程不用你会写 JavaScript。但你得会描述"哪里不对"：它做出来的第一版一般能玩，后面每一处手感、画面、规则，都要靠你一句句说。