メニュー
Capture and analyze Imla Android renderer smoke traces with the repo-local Perfetto config, SQL queries, and Android Performance Analyzer workflow.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/desugar-64/imla --skill imla-perfetto-feedbackコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
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.
SOC 職業分類に基づく
SKILL.md を表示中
| 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.