菜单
Capture and analyze Imla Android renderer smoke traces with the repo-local Perfetto config, SQL queries, and Android Performance Analyzer workflow.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/desugar-64/imla --skill imla-perfetto-feedback命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Capture Imla Android screenshots through the repo-local half-size ADB screenshot wrapper to reduce visual-inspection token cost while preserving enough detail for renderer checks.
Run Imla ADB commands through the repo-local timeout wrapper so physical-device installs, launches, input, screenshots, package-manager calls, and diagnostics fail fast instead of hanging.
Read Imla scene renderer counters from repo-local Perfetto analysis. Use when evaluating render reasons, root captures, slot content captures, mask/clip captures, geometry refreshes, scene renders, or whether an interaction is content-capture-heavy versus geometry-only.
基于 SOC 职业分类
正在显示 SKILL.md
| name | imla_perfetto_feedback |
| description | Capture and analyze Imla Android renderer smoke traces with the repo-local Perfetto config, SQL queries, and Android Performance Analyzer workflow. |
Use this skill before and after renderer, blur, GL threading, capture, or Compose modifier changes that need runtime feedback beyond screenshots.
The repo-local tool is:
tools/imla-perfetto-feedback
It wraps the checked-in Perfetto smoke config, named SQL queries, screenshot capture, and optional Android Performance Analyzer opening. The extensionless shell wrapper is the stable entrypoint; the .py file is the implementation detail.
Analyze an existing trace:
tools/imla-perfetto-feedback analyze diagnostics/apa/traces/imla-smoke-2026-05-23.perfetto-trace
Capture a new smoke trace on the active emulator, exercise the demo, pull a screenshot, analyze it, and open it in APA:
tools/imla-perfetto-feedback smoke --device emulator-5580 --open-apa
Capture without scripted input:
tools/imla-perfetto-feedback capture --device emulator-5580 --no-exercise
Open a trace in Android Performance Analyzer:
tools/imla-perfetto-feedback open diagnostics/apa/traces/<trace>.perfetto-trace
Start investigation from these labels when present:
ImlaSceneCoordinator#commitRootCaptureImlaSceneCoordinator#refreshSlotGeometryRenderableRootLayer#captureLayerGraphicsLayerTexture#captureGraphicsLayerGraphicsLayerTexture#captureCanvasSceneLayerRepository#captureSlotContentSceneMaskRepository#captureSlotMasksSceneGlRenderer#renderRenderer2D#endSceneRenderer2D#uploadVerticesvboSetDataglBufferDataKey threads are the main app thread, Android RenderThread, and Imla GLUiLayerRender.
./gradlew -q :app:assembleDebug then install or launch via ADB.tools/imla-perfetto-feedback smoke --device <serial>.perfetto-trace-analysis and perfetto-sql for deeper AI-assisted trace questions; this repo tool provides the repeatable capture/query baseline.The scene counter summary is written to 60_scene_counters.txt when scene counters are present.