简体中文说明:README.zh-CN.md
Beta release candidate for agentic WeChat Mini Program visual debugging inside WeChat DevTools / 微信开发者工具.
Status: public GitHub beta for real Mini Program development workflows with documented smoke runs.
wechat-devtools-automator gives agents CLI-powered eyes on real Mini Program pages: run preflight, compile when needed, list routes, open a page with query params, click or type to reach the right state, scroll below the fold, capture screenshots, and collect console / exception output as evidence.
- Distribution posture today: public GitHub beta for evaluation and smoke-tested adoption
- License posture today: MIT
- Release trust anchors: documented smoke runs, packaged
.skillartifact, trigger QA summaries, and sharable evidence folders
Before promoting the repo more broadly, point readers to LICENSE, references/install-and-share.md, assets/release-checklist.md, and references/github-metadata.md.
- A route-aware evidence folder under
<project>/output/wechat-devtools-automator/<run-id>/ - Page screenshots and optional GUI crop
report.jsonwith route, query, action, console, and exception context- A reproducible CLI trail that another agent or reviewer can rerun
Mini Program frontend work often fails at the “eyes-on” step. Browser projects can lean on Playwright, but Mini Program UI verification usually passes through WeChat DevTools. If the agent cannot see the real rendered page, many judgments regress into guesswork.
This repo turns that into a repeatable workflow:
doctorto verify the local environmentroutesto discover available pagesshot/scroll-shot/gui-shotto capture real rendered UI--tap,--longpress,--input,--trigger,--actionto reach post-interaction statesreport.jsonoutput withconsole/ exception evidence
The packaged .skill targets Codex today, but the core workflow is just the wrapper plus Node scripts.
If your agent tool can run local shell commands, you can use the same flow from Codex, Claude Code, Cursor, OpenCode, OpenClaw, or similar agent-driven setups by calling:
scripts/wechat_devtools_automator.shClone the repo anywhere convenient, then call the wrapper directly:
git clone https://github.com/YogurtJ/wechat-devtools-automator.git
cd wechat-devtools-automator
export WDA="$(pwd)/scripts/wechat_devtools_automator.sh"
"$WDA" doctor --project /path/to/mini-program-project
"$WDA" routes --project /path/to/mini-program-projectexport CODEX_HOME="${CODEX_HOME:-$HOME/.codex}"
mkdir -p "$CODEX_HOME/skills"
git clone https://github.com/YogurtJ/wechat-devtools-automator.git \
"$CODEX_HOME/skills/wechat-devtools-automator"
export WDA="$CODEX_HOME/skills/wechat-devtools-automator/scripts/wechat_devtools_automator.sh"
"$WDA" doctor --project "$(pwd)"
"$WDA" routes --project "$(pwd)"Prefer a packaged .skill instead? See references/install-and-share.md.
- MIT license lowers adoption friction for teams and tool builders
- Works as a normal shell-driven repo even outside Codex
- Keeps the Mini Program feedback loop grounded in real WeChat DevTools output
- Produces shareable evidence instead of “trust me, it rendered fine”
export WDA="/path/to/wechat-devtools-automator/scripts/wechat_devtools_automator.sh"
"$WDA" doctor --project "$(pwd)"
"$WDA" routes --project "$(pwd)"
"$WDA" shot --project "$(pwd)" --route "<copy-a-route-from-routes-output>"Artifacts default to:
<project>/output/wechat-devtools-automator/<run-id>/
After you have one real route from routes, continue with:
"$WDA" shot --project "$(pwd)" --route "<real-route>" --query id=item_001
"$WDA" shot --project "$(pwd)" --route "<real-route>" --tap "<selector>"
"$WDA" shot --project "$(pwd)" --route "<real-route>" --input "<selector>::奶茶"
"$WDA" scroll-shot --project "$(pwd)" --route "<real-route>" --scroll-step 900 --scroll-captures 3
"$WDA" gui-shot --project "$(pwd)" --route "<real-route>"Need a safer first-run path? See references/quickstart.md.
- “帮我在微信开发者工具里打开这个小程序页面并截图。”
- “先编译一下,再带参数打开指定页面让我看真实渲染。”
- “这个页面得先点一下 tab 才能看到目标状态,帮我点开再截图。”
- “帮我往下滚几屏,把首屏以下也截出来。”
- “打开页面后把 console 报错也一起抓出来。”
- “Open a WeChat Mini Program route with query params and capture screenshots.”
- UI validation for Mini Program pages
- Visual debugging after route/query changes
- Interaction-driven states that only appear after tapping or typing
- Below-the-fold inspection
- Collecting screenshots and console evidence for agent handoff or bug reports
- Generic website browser automation
- Whole-desktop screenshots unrelated to WeChat DevTools
- Pure code refactors that do not require visual verification
- A claim of native integration with every agent IDE; outside Codex, the wrapper/script path is the supported route
- Depends on DevTools GUI/session state; parallel runs on one machine can contend
- Visual results can vary by local font, OS, zoom, and mock / API data
gui-shotis macOS-first; simulator/page screenshots are the portable default- If
build-npmfails due to missing npm deps in the project, fix project deps first
| Platform | Node | WeChat DevTools | Validation |
|---|---|---|---|
| macOS 15.5 | Node v24.13.1 | 2.01.2510280 | doctor + routes + shot + scroll-shot on 2026-03-24 |
Record every newly verified platform combo in assets/release-checklist.md.
- For first-time users, link this README plus
README.zh-CN.mdandreferences/quickstart.md - For release QA, include
assets/release-checklist.mdand the trigger QA summary - For handoff, always share the screenshot folder and
report.json, not screenshots alone - If you publish a GitHub repo, fill the repo About box/topics using
references/github-metadata.md
- Chinese README:
README.zh-CN.md - Install and share:
references/install-and-share.md - Trigger QA:
references/eval.md - Release metadata:
references/github-metadata.md - License:
LICENSE