Przejdź do treści
Biblioteka skilli
Programowanie · Analiza

recreate-perf

/recreate-perf

Ustala i pilnuje budżetu wydajności klonu, mierzy Core Web Vitals i rozmiar bundla względem oryginału oraz podpowiada poprawki.

Repozytorium
recreate ↗
Licencja
MIT
Ostatnia zmiana repozytorium
05.11.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
1
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
1
Przestarzałość

Brak sygnałów. Repozytorium zmienione 05.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Ustala i pilnuje budżetu wydajności klonu, mierzy Core Web Vitals i rozmiar bundla względem oryginału oraz podpowiada poprawki.

Wyzwalacze z opisu

faster than Xs slowperformanceCore Web VitalsLighthousebundle sizemake it fasterspeed compared to the original
Oryginalny opis (EN)

Sets and enforces a performance budget for the clone, and measures it against the original so "faster than X" is a number you can defend. Core Web Vitals, bundle size, request count, a budget in CI, and the fixes in the order that actually pays. Use when the user says "it's slow", "performance", "Core Web Vitals", "Lighthouse", "bundle size", "make it faster", "speed compared to the original", or after /recreate-build.

Jak działa

  1. Step 1: measure the original first
  2. Step 2: set the budget
  3. Step 3: measure yours the same way
  4. Step 4: fix in the order that pays
  5. Step 5: hold the line in CI
  6. Step 6: the claim
  7. Output
Sekcje z nagłówków SKILL.md, w oryginale · 133 linii, 840 słów
Podgląd SKILL.md
---
name: recreate-perf
description: >-
  Sets and enforces a performance budget for the clone, and measures it
  against the original so "faster than X" is a number you can defend. Core Web
  Vitals, bundle size, request count, a budget in CI, and the fixes in the
  order that actually pays. Use when the user says "it's slow", "performance",
  "Core Web Vitals", "Lighthouse", "bundle size", "make it faster", "speed
  compared to the original", or after /recreate-build.
---

# recreate-perf

"It's slow" is one of the most common complaints `/recreate-listen` finds.
If it is true of the original, speed is not a nice-to-have for you, it is
the product difference, and it is the easiest one to prove. This stage makes
it measurable instead of asserted.

Reads `recreate/feedback.md` if the listen stage has run. Writes
`recreate/perf.json` and `recreate/perf.md`.

```bash
python3 recreate/u.py perf recreate/perf.json
python3 recreate/u.py perf recreate/perf.json --lighthouse lh.json --route / --write
```

## Step 1: measure the original first

Do this before you optimise anything, because once you have shipped your own
version you will never get a clean baseline again.

Run Lighthouse against the original's main public pages, its home page, its
pricing page, and whatever public page its core flow starts on. Mobile
preset, throttled, three runs, take the median. Record **how** and **when**
in `perf.json`:

```json
"baseline": {
  "measured": "2026-11-05",
  "how": "Lighthouse 12, mobile preset, simulated 4G, median of 3",
  "/": { "lcp_ms": 3900, "cls": 0.21, "tbt_ms": 640 }
}
```

Measure only public pages you can reach as a normal visitor. Three runs, not
a load test, never hammer the original's servers.

## Step 2: set the budget

Defaults, which are the thresholds Core Web Vitals calls good:

| metric | budget |
| --- | --- |
| Largest Contentful Paint | 2500 ms |
| Interaction to Next Paint | 200 ms |
| Cumulative Layout Shift | 0.1 |
| Time to First Byte | 800 ms |
| Total Blocking Time | 300 ms |
| JavaScript transferred | 300 kB |
| CSS transferred | 60 kB |
| Total transferred | 1000 kB |
| Requests | 50 |

Tighten any metric where the original is bad and you intend to say so. If
their LCP is 3.9s, a budget of 2.5s is not ambitious enough to be a story;
1.5s is.

A budget is per route, and the routes that matter are the ones in the core
flow from `recon.json`, plus the landing page. Do not budget every page.

## Step 3: measure yours the same way

Same Lighthouse version, same preset, same throttling, same number of runs,
against a production build, never the dev server, which is slower in ways
that tell you nothing.

```bash
npx lighthouse https://localhost:3000 --output json --output-path lh.json \
  --preset=desktop --quiet
python3 recreate/u.py perf recreate/perf.json --lighthouse lh.json --route / --write
```

The tool prints every metric against its budget, and where a baseline exists,
how far ahead or behind the original you are.

## Step 4: fix in the order that pays

Work down this list. Most clones find their whole problem in the first three.

1. **Ship less JavaScript.** Check what is in the bundle before guessing.
   The usual culprits are a date library, an icon set imported whole, a chart
   library on a page with no chart, and a UI kit pulled in for two components.
2. **Server-render the first screen.** If the core flow's first paint waits
   on a client fetch, that is the LCP, and no amount of tuning fixes it.
3. **Images.** Correct dimensions set, modern format, lazy below the fold,
   eager and preloaded for the LCP image.
4. **Fonts.** Subset, preload the one weight above the fold, `font-display:
   swap`, and a metric-matched fallback so the swap does not shift layout.
5. **Reserve space.** Every image, embed and ad slot gets explicit dimensions.
   Most CLS is one unreserved element.
6. **Database.** The indexes the architect stage specified, actually created.
   One missing index on the core flow's main query outweighs every frontend
   fix on this list.
7. **Cache.** Static assets immutable and far-future; HTML at the edge where
   the page allows it.

Re-measure after each change rather than at the end. You want to know which
one worked.

## Step 5: hold the line in CI

A budget nobody enforces drifts back within a month. Add the check to the
pipeline so a pull request that blows the budget fails:

```yaml
- run: npm run build
- run: npx lighthouse-ci autorun   # or your own Lighthouse run
- run: python3 recreate/u.py perf recreate/perf.json
```

The tool exits 1 when any measured metric is over budget.

## Step 6: the claim

If your main page beats theirs by a wide margin, that is a landing-page
asset. Use it honestly or not at all:

- State the number, the date, and how it was measured.
- Compare like with like: same page type, same device profile, same network.
- Re-measure before each time you repeat it. They ship too, and a stale claim
  is a false one.
- Never compare your local build against their production site.

The tool will not print a headline under a 10% difference, because below that
the measurement noise is larger than the claim.

## Output

`recreate/perf.json` (budgets, per-route measurements, baseline with its
method and date), `recreate/perf.md` (what was slow, what was changed, what
it bought), the CI check wired in, and the honest comparison line if the
evidence supports one. Record the stage.

Wersje

W bazie nie ma innych skilli o nazwie recreate-perf.

Migawka bazy z 13.11.2026 · wersja paczki e3f2364d007f (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 90perf-audit
    Audytuje rozmiar bundla frontendu i ryzyka Core Web Vitals, aby znaleźć wąskie gardła po stronie przeglądarki.
  • 100perf-web-optimization
    Optymalizuje wydajność stron WWW: rozmiar bundle'a, obrazy, cache, lazy loading, przesunięcia układu i czas do interaktywności pod wynik Lighthouse.
  • 100web-performance
    Diagnozuje wolne strony i Core Web Vitals, mierząc Lighthouse i analizę bundle, a potem stosuje celowane poprawki.
  • 100frontend-performance
    Audytuje wydajność webową głębiej niż Lighthouse: Core Web Vitals, rozmiar bundle, kompresję, cache, HSTS, resource hints, service worker i speculation rules.
  • 100performance-profiling
    Diagnozuje wolną stronę lub aplikację, mierząc Core Web Vitals, Lighthouse, rozmiar pakietów i zużycie pamięci.
  • 100fec-performance-optimization
    Diagnozuje i poprawia wydajność frontendu: Core Web Vitals, rozmiar bundla, koszt renderowania, wodospady sieciowe, wycieki pamięci, długie zadania, wyniki Lighthouse i budżety wydajności.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.