Perfetto Performance Analysis
Análisis de trazas Perfetto con un trace_processor fijado y 235 gráficos de consulta
Probado · Funciona
Qué hace
Analiza trazas Perfetto de Android, Linux y Chromium ejecutando SQL determinista contra un checksum-pinned trace_processor_shell instead of guessing from symptom names. Bundles 235 runnable analysis graphs plus 14 workflow guides covering startup, jank y scrolling, input latency, ANR y binder blocking, CPU scheduling, memory y GC, GPU y SurfaceFlinger, power y thermal. Se activa cuando se le entrega a Claude un .pftrace o perfetto-trace y se pregunta por qué una app es slow, janky, o regressed between two builds.
Informe de la prueba
Arrancó su propio Perfetto v57.2 trace_processor y verificó el SHA-256 del binary contra el bundled lock file, then ran startup_analysis end to end on a real 2 MB Android warm-start trace en 13.7 seconds. One command returned a 15-step evidence graph: un 156.5 ms warm start de androidx.benchmark.integration.macrobenchmark.target, activityResume holding 68% of it con only 2.2 ms de self time, 11 synchronous main-thread binder calls into system_server (14.8% de startup), 89.8% del main thread asleep, y una data-quality flag (R008_TTFD_GT_DUR) on the TTFD figure. El honest no-skill baseline, hand-written SQL through the official perfetto Python library, found the package name y un 144 ms launching slice but no self-time, thread-state o binder attribution y no quality warning. El bundled SQL guardrail also rejected a bare span_join call on the spot, which is exactly the Perfetto footgun it advertises. Only rough edge: some result strings come back in Chinese despite English-only documentation.
Probado el: 2026-08-10 · Claude Code 2.x (agent harness)
Instalación
git clone --depth 1 https://github.com/Gracker/Perfetto-Skills.git mkdir -p ~/.claude/skills cd Perfetto-Skills && cp -r skills/perfetto-performance-analysis ~/.claude/skills/perfetto-performance-analysis
Comandos y prompts de ejemplo
/perfetto-performance-analysisAnálisis de trazas Perfetto con un trace_processor fijado y 235 gráficos de consulta
Los skills se activan con peticiones en lenguaje natural, sin comandos que memorizar. Tras instalarlo, prompts como estos lo activan (en inglés):
Analyze this Perfetto trace and find the app's startup durationWhat's causing jank and scroll latency in this trace file?Find the top main-thread blocking contributors in this trace