Miniapp Precision Control Evolution

Problem

Mini-app projects often fail not because code is missing, but because edits are applied to the wrong target: global replacement hits duplicate styles, rpx is confused with px, setData receives large objects, scoped styles leak, WeChat review rules are missed, or a build succeeds while the real device still shows old content. This skill treats those changes as controlled engineering work for WeChat, Alipay, Baidu, Douyin, and uni-app projects.

How it works

• Mode first: it defaults to strict mode and only changes the file, property, and position explicitly requested; extra polish is not added unless explicitly authorized.
• Intent parsing: natural requests such as “reduce speed by one third” are translated into parameter direction, with the current value checked first and multiple needs separated before editing.
• Unique targeting: it prefers line numbers, full CSS selectors, class/id anchors, and surrounding context instead of global replacement or fuzzy matching for repeated text.
• Verification loop: after a change, it reads back the result, rebuilds, and confirms the effect in the devtools or on a device; JSON edits use Node-based read/write to avoid BOM-related parse issues.
• Failure circuit breaker: after two failures in the same direction, it stops repeating the same action and uses project-wide search or asks for the correct target page.

Scope

It fits style tweaks, copy changes, localized logic fixes, and pre-release checks in WeChat mini-apps, uni-app, and other mini-program runtimes. It is especially aware of platform constraints such as rpx conversion, path-based setData, review red lines, subscription messages, payment callbacks, cloud-development permissions, and the rule that secrets like AppSecret must not be stored on the frontend. It is not a general web development substitute, and large refactors should still be broken into multiple verified commits.