perfetto
Installation
SKILL.md
Perfetto
PATHS: resolve bundled resources from the activated skill directory, not the target project's working directory. This package supplies recipes, not installed Perfetto binaries or a device connection.
1. Select the signal and boundary
- RECORD the question, target process/thread, scenario, device/OS, time interval, existing trace path, and tool version. Separate the capture target from the analysis host. Preserve originals and capture configuration.
- Existing trace -> READ references/analysis.md. Discover recorded signals before selecting an analysis; missing scheduler, samples, heap graph, or FrameTimeline data cannot be recovered by a query.
- New Android/Linux/Chrome capture -> READ references/recording.md. Choose only required sources and a bounded reproduction window. Obtain authorization before elevated capture, app restarts, heap dumps, or changes to device/security settings; traces can contain personal data, paths, arguments, and object contents.
- In-app events -> READ references/instrumentation.md. Prefer existing Android Trace/ATrace annotations where sufficient, C++ TrackEvents for portable application instrumentation, custom sources only for data that needs them.
- Tool acquisition or Perfetto source changes -> READ references/tools.md. Match syntax to actual
--version/--help: this baseline is v59.1, not a promise that an Android system binary or installed Python package has the same commands. - Keep scope explicit: built-in system probes cover Android/Linux; SDK/UI/analysis also support desktop platforms. Perfetto is client-side tracing, not an OpenTelemetry-style distributed recording service.
DONE: the required evidence, supported capture route, tool version, and sensitive/privileged boundaries are explicit.