How do I install the Perfsweep skill?

Run npx skillmds@latest add testbeagle/perfsweep in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Perfsweep safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Perfsweep?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Perfsweep free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Perfsweep?

TestBeagle (@testbeagle) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Perfsweep by testbeagle — Agent Skill

Name: Perfsweep
Author: TestBeagle

Perfsweep

Use when asked to test performance of a locally runnable app — Lighthouse, load time, jank, bundle size, low-power/energy use. Triggers on "성능 점검", "Lighthouse", "느린 이유", "번들 크기", "저전력".

TestBeagle ee02c6e 2.7 KB Updated Sep 17, 2026

File contents

perfsweep

Plan-gated performance audit of a locally running app: load, runtime jank, bundle weight, and energy use — measured, not guessed, with numbers tied to a trace.

Phase 0 — Discover (reuse preflight)

Run/reuse preflight for targets, run commands, drivers, output location. Reuse any existing perf budgets or bundle-analysis config in the repo.

Phase 1 — Plan + approval gate (MANDATORY)

Present the routes/scenarios to measure and the metrics per platform, then stop per the shared gate in ../beagle-shared/approval-gate.md. Note that numbers are dev-build/local, not production figures.

Phase 2 — Measure

Web (../beagle-shared/drivers-web.md):

Lighthouse per key route (performance category) — LCP, CLS, TBT, TTI: npx lighthouse http://localhost:PORT/route --only-categories=performance --output=json --quiet --chrome-flags="--headless=new" (any driver), or chrome-devtools MCP lighthouse_audit.
Interaction trace: agent-browser trace start → interact → agent-browser trace stop OUT/trace.json, or chrome-devtools performance_start_trace → performance_stop_trace → performance_analyze_insight (only the MCP summarizes insights for you; with a raw trace, read long tasks from the JSON or mark the insight 검증 불가).
Console long-task / excessive re-render warnings after each route (agent-browser console). Capture-only driver → interaction metrics are 검증 불가.
Bundle weight (any driver): run the repo's build + bundle analyzer; flag oversized entries and heavy dependencies.

Mobile (../beagle-shared/drivers-mobile.md — Performance section):

Cold-start: Android adb shell am start -W (read TotalTime); iOS xctrace/Instruments where available, else 검증 불가.
Dropped frames / jank: Android adb shell dumpsys gfxinfo <pkg>.
Low-power / energy: idle CPU and continuous work — flag idle animations, polling, always-on computation (Android dumpsys batterystats / top; iOS Instruments Energy where available). Where the tools at hand don't expose a metric, mark it 검증 불가.

Phase 3 — Report

Write per ../beagle-shared/report-base.md (Korean default, anti-slop). Per finding: 지표 · 측정값(+기준선/버짓 대비) · 위치 · 왜(검증: trace/Lighthouse 증거) · 수정 제안 · 확신. Measure before recommending — no "optimize X" without a number showing X is the cost. Unmeasured scenarios → 검증 불가.

Optionally offer a static re-measure runner via ../beagle-shared/emit-runner.md.

---name: perfsweepdescription: Use when asked to test performance of a locally runnable app — Lighthouse, load time, jank, bundle size, low-power/energy use. Triggers on "성능 점검", "Lighthouse", "느린 이유", "번들 크기", "저전력".---# perfsweepPlan-gated performance audit of a locally running app: load, runtime jank, bundle weight, and energy use — measured, not guessed, with numbers tied to a trace.## Phase 0 — Discover (reuse preflight)Run/reuse `preflight` for targets, run commands, drivers, output location. Reuse any existing perf budgets or bundle-analysis config in the repo.## Phase 1 — Plan + approval gate (MANDATORY)Present the routes/scenarios to measure and the metrics per platform, then stop per the shared gate in `../beagle-shared/approval-gate.md`. Note that numbers are dev-build/local, not production figures.## Phase 2 — Measure**Web** (`../beagle-shared/drivers-web.md`):- Lighthouse per key route (performance category) — LCP, CLS, TBT, TTI: `npx lighthouse http://localhost:PORT/route --only-categories=performance --output=json --quiet --chrome-flags="--headless=new"` (any driver), or chrome-devtools MCP `lighthouse_audit`.- Interaction trace: `agent-browser trace start` → interact → `agent-browser trace stop OUT/trace.json`, or chrome-devtools `performance_start_trace` → `performance_stop_trace` → `performance_analyze_insight` (only the MCP summarizes insights for you; with a raw trace, read long tasks from the JSON or mark the insight 검증 불가).- Console long-task / excessive re-render warnings after each route (`agent-browser console`). Capture-only driver → interaction metrics are 검증 불가.- Bundle weight (any driver): run the repo's build + bundle analyzer; flag oversized entries and heavy dependencies.**Mobile** (`../beagle-shared/drivers-mobile.md` — Performance section):- Cold-start: Android `adb shell am start -W` (read TotalTime); iOS `xctrace`/Instruments where available, else 검증 불가.- Dropped frames / jank: Android `adb shell dumpsys gfxinfo <pkg>`.- **Low-power / energy**: idle CPU and continuous work — flag idle animations, polling, always-on computation (Android `dumpsys batterystats` / `top`; iOS Instruments Energy where available). Where the tools at hand don't expose a metric, mark it 검증 불가.## Phase 3 — ReportWrite per `../beagle-shared/report-base.md` (Korean default, anti-slop). Per finding: 지표 · 측정값(+기준선/버짓 대비) · 위치 · 왜(검증: trace/Lighthouse 증거) · 수정 제안 · 확신. Measure before recommending — no "optimize X" without a number showing X is the cost. Unmeasured scenarios → 검증 불가.Optionally offer a static re-measure runner via `../beagle-shared/emit-runner.md`.

Frequently asked questions