snapshot-testing
/snapshot-testing
Konfiguruje testy snapshotów z wzorcami w projekcie Arkive, diagnozuje nieudane weryfikacje i wiąże je z CI.
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 23.09.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
Co robi i kiedy
Konfiguruje testy snapshotów z wzorcami w projekcie Arkive, diagnozuje nieudane weryfikacje i wiąże je z CI.
Oryginalny opis (EN)
Use when: (1) enabling golden snapshot testing in a project that uses Arkive, (2) a verifyShowcase build failed and needs diagnosing, (3) the UI changed intentionally and goldens need re-recording, (4) wiring snapshot verification into CI. There is no test to write — Arkive generates it; this skill configures retention, records goldens, proves verification works, and diagnoses failures. Requires arkive 0.0.3+.
Jak działa
- Step 1 — Retention
- Step 2 — Pre-flight the repo
- Step 3 — Record and confirm the goldens actually landed
- Step 4 — Prove verification works before committing
- Step 5 — Diagnosing a failed verify
- CI
- Red flags — STOP
Podgląd SKILL.md
---
name: snapshot-testing
description: 'Use when: (1) enabling golden snapshot testing in a project that uses Arkive, (2) a verifyShowcase build failed and needs diagnosing, (3) the UI changed intentionally and goldens need re-recording, (4) wiring snapshot verification into CI. There is no test to write — Arkive generates it; this skill configures retention, records goldens, proves verification works, and diagnoses failures. Requires arkive 0.0.3+.'
---
# Arkive Snapshot Testing
Lock in approved UI with goldens and make regressions fail the build via
`verifyShowcase<Variant>`. Read
`references/arkive-cheatsheet.md` for retention semantics, paths,
and failure behavior.
**Scope:** this is code-vs-code-as-last-approved, judged pixel-exact against committed
goldens. Code-vs-*design* (Figma/specs, judged by eye) is `/arkive:design-loop` — it
needs no goldens at all. Don't conflate them.
**Version gate (do this first):** check the project's pinned arkive version.
`verifyShowcase` exists from **0.0.3**; on anything older, verification silently cannot
fail. If the pin is older, upgrade first (`/arkive:setup` covers resolving latest).
## Step 1 — Retention
Read the module's `arkive { snapshotRetention }`:
- **If NONE (or unset — NONE is the default): set it to `BASE`, and explain the change to
the user, don't just do it silently.** The explanation: under NONE the showcase
consumes every snapshot and nothing is verifiable; BASE keeps exactly one golden per
component, which is the regression signal, without committing every font/density/RTL
variant — a real app went from 57 goldens to 725 with variants retained, which is why
ALL is not the default recommendation.
- **If BASE or ALL already:** leave it alone; the team made a choice.
- Suggest `ALL` only if the user explicitly wants variant-level regression coverage, and
pair it with Git LFS for the golden directory.
## Step 2 — Pre-flight the repo
Check `.gitignore` (root and module) for the golden directory
(`src/test/snapshots/`). Goldens must be **committable** — verification in CI is
meaningless against images that only exist on one laptop. If ignored, surface it and fix
it with the user (scoped: Arkive's files are all prefixed `com.infinum.arkive_`).
## Step 3 — Record and confirm the goldens actually landed
Run `./gradlew :<module>:generateShowcase<Variant>` (recording takes minutes), then
verify the outcome — do not assume it:
1. Count `com.infinum.arkive_`-prefixed PNGs in `<module>/src/test/snapshots/images/`.
2. Cross-check against the component count in the module's `arkive-showcase.json` —
under BASE they should match one-to-one.
3. Scan the build log for `Arkive: no snapshot recorded for component` warnings. A
crashed preview has **no golden and will never be covered by verification** — report
each one to the user rather than letting coverage silently shrink. (Re-run with
`--info` to see the test-side crash message.)
## Step 4 — Prove verification works before committing
Run `./gradlew :<module>:verifyShowcase<Variant>` immediately. It must be green against
the goldens just recorded. Only then commit the goldens. Committing unverified goldens
means the first CI failure is undebuggable ("was it ever green?").
## Step 5 — Diagnosing a failed verify
`verifyShowcase` fails with **one aggregate error naming every mismatched component**,
each with a delta-image path
(`<module>/build/paparazzi/failures/<variant>/delta-*.png`) and an accept command.
| Failure shape | Meaning | Action |
|---|---|---|
| `Arkive: snapshot verification failed for N component(s)` | Rendering differs from the golden | **View the delta image** for each named component, then decide: regression → fix the code; intentional change → re-record |
| Missing golden for a component | Component added/renamed since goldens were recorded | Re-record and commit the new golden |
| `verifyShowcase... has nothing to verify — snapshotRetention is NONE` | Fails fast at configuration time | Go to Step 1 |
| The consumer's own `verifyPaparazzi` failing on non-Arkive tests | Their tests, their goldens | Out of scope — Arkive's generated test self-skips what retention didn't keep and never affects theirs |
**Intentional change flow:** `generateShowcase<Variant>` (re-records everything and
retention re-prunes) → `verifyShowcase<Variant>` green → commit the updated goldens in
the same PR as the UI change.
## CI
Run `verifyShowcase<Variant>` on pull requests. Goldens live in the repo; **recording
happens locally, never in CI** — a CI that re-records and verifies in one run can never
fail. If JVM rendering differs across OSes (fonts), pin the CI runner OS to what the
team records on.
## Red flags — STOP
- Re-recording to make a failing verify pass **without viewing the delta image first**.
"Accept the changes" is for intentional changes; a regression accepted is a regression
shipped.
- Committing goldens that were never proven green (Step 4 skipped).
- Flipping retention to ALL casually — 10× golden growth is a repo decision.
- Treating a crashed preview's absence from verification as "covered".
- Declaring snapshot testing "set up" when the goldens directory is still gitignored.
Pliki w paczce (2)
- SKILL.md 5 KB
- references 1 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · infinum/arkive | 90 | Apache-2.0 |
| jacob-balslev/skill-graph | 100 | Apache-2.0 |
| Amey-Thakur/AI-SKILLS | 90 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki ed890faa05d8 (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
- 100setup
Dodaje bibliotekę Arkive do projektu: instaluje wersję z Maven Central, konfiguruje moduły z podglądami i generuje pokaz. - 90design-loop
Wdraża interfejs z projektu w Figmie w aplikacji Arkive i porównuje wynik z projektem, aż różnice zostaną usunięte. - 100golden-and-snapshot-tests
Pilnuje testów golden i snapshotów danych oraz pikseli: pliki *.golden.test.ts, wpisy kontraktowe tylko do dopisywania, tolerancja Skia i zasady użycia jest -u. - 100run-goldens-rebaseline
Przeprowadza rytuał aktualizacji wzorców w Jest: najpierw asercje, potem przegląd każdej zmiany i usunięcie zbędnych plików. - 100snapshot-test-setup
Konfiguruje testy regresji wizualnej SwiftUI z swift-snapshot-testing, generuje szablon testów snapshotowych i konfigurację CI, aby zapobiegać regresjom interfejsu. - 100managing-snapshot-tests
Tworzy i waliduje migawki komponentów do testów regresji interfejsu użytkownika, w tym aktualizację snapshotów.