AI-Assisted Rework of WeChat Mini-Game Uses Automated Playtesting
A developer reworked the WeChat mini-game now called 合合兵团, retaining only its auto-chess mode, removing bonds, adding two factions and hidden units, and redrawing 40 unit images. The developer used AI to turn subjective feedback into concrete rules, diagnose a performance problem, and run 8–16 automated matches after rule changes to assess difficulty. Claude Code was used to operate ChatGPT in a browser for image generation. In a self-reported 16-run simulation, 7 of 8 runs cleared layers 1–9, while 7 of 8 lost once at the layer-10 boss. The developer said creative decisions such as the game’s direction, name, and visual direction remained human choices.
The report documents a specific workflow in which the developer made creative decisions while using AI to translate feedback into rules, generate images, identify the cause of a performance problem, and support automated difficulty testing. The playtest results were self-reported and not independently verified.
上图是游戏现在的样子:名字从「圣光防线」改成了「合合兵团」,首页只剩一个「开始游戏」,进去先选阵营,合到最高星会出隐藏兵种。上一篇预告这一篇讲资质手续,手续还在办,软著也还在等答复,所以这一篇先讲 11 月 2 日到 8 日这一周的主要工作:改玩法。
这一周做了五件事:查到了上一篇遗留的卡顿的根因;只保留自走棋闯关一个模式,地图改成无尽;把「没有成就感、没有危急感」改成六条具体规则;定下以兵种为主线,改名,去掉羁绊,分成两个阵营并各加一个隐藏兵种;按一份画风规格重画了 40 张兵种图,加了开场故事。一周下来测试从 737 个增加到 828 个,微信包 16.13 MB。