Phaser 4 Testing and Performance

Global Control Access

Control Plane Boundaries: Proposals, reviews, and modifications are allowed within the scope of Work Item task authorization or explicit approval, and must return to the

$phaser4-game-workflow-control

risk gate.

This domain can propose verification plans and review candidates, and can only write tests or evidence within established, validly authorized Work Items, Level A and paths; results are returned to

phaser4-game-workflow-control

for audit and state migration. Test passing cannot cover failures in authorization, scope, path, baseline or A4-A6 operation approval.

Evidence takes precedence over subjective conclusions. When the project has not defined quantitative indicators, only report actual measurements and decision gaps, and do not create universal thresholds on your own.

Read Work Item, Quality Gate, TDD, qa-plan and current candidate. F0 verifies authorization and process, F1 verifies specification consistency, F2 forms independent QA domain conclusions, F3 binds actual commands, environment, data sources, files, judgments and uncovered items; mark as unverified if independent review is missing.
Check the same project, service type, mode, port and health status before starting the service; reuse existing healthy instances and do not terminate processes with unknown ownership.
Visual tasks check V0 diversion and V1/V2 condition gates. When the Work Item specifies a render as the reconstruction target, QA checks the frozen visual target, comparison conditions, per-state/per-region fidelity matrix and approved exceptions according to Visual Reconstruction; adaptations within tolerance that do not change visual facts can be marked as
```
AUTO
```
, any visible deviation or substantial trade-off only requests precise confirmation once. Professional fixes do not automatically authorize changes to frozen visual facts.
When V1 applies, check visual targets, information hierarchy, sketches/gray boxes, interactions, layouts, failure recovery and budget. Faithful reconstruction is recorded with
```
AUTO
```
decision basis and proceeds to V2 only when there are no visible deviations or substantial trade-offs, and engineering adaptations are within the project's predefined tolerance; otherwise, request precise selection once, record
```
USER_DECISION
```
and approved exceptions. Atomic resources are bound to the current
```
AUTO
```
or
```
USER_DECISION
```
record.
Common assets/layouts allow
```
not-applicable
```
for reconstruction. Renders in
```
v3-ready
```
check targets, contract alignment, coverage with bounds and condition number evidence; V3 must also record explicit
```
production_method
```
,
```
delivery_kind
```
,
```
image_generation_required
```
,
```
generation_record_required
```
,
```
substitution_policy
```
and
```
expected_assets
```
per
```
annotation_number/region_id
```
. The
```
visualProductionUnits
```
of V3 implementation packages must map one-to-one with coverage, and output paths, ownership and formats must not conflict.
V4 must submit
```
production_contract_audit
```
, checking expected vs actual methods, delivery types, output files, generation/prompt records and runtime consumption per region; F2 only accepts machine-verified facts with
```
validationMode=MACHINE
```
for the current identity. Only V5
```
complete/verified
```
requires runtime measurement, F3 runtime replay, freshness-bound fidelity cases, actual consumption and no unapproved substitutions; any identity change rejects old evidence, and no repeated visual review is required after V2's only manual confirmation passes.
Screenshots under the same conditions record viewport, device pixel ratio, state, trajectory, language, random seed, ROI, project tolerance defined before implementation, dynamic time sampling/stable frames and masking reasons. Frozen target viewport/state takes the complete viewport as primary evidence; ROI, side-by-side, overlay or pixel difference are only supplementary; other viewports verify visual intent and relationship invariants according to layout contracts, and generative content, animations and VFX cannot rely solely on pixel difference. Unexplained or out-of-tolerance differences, lack of reference/candidate evidence under the same conditions, lack of approved exceptions or conclusions based solely on subjectivity are all judged as failed.
UI reads the
```
phaser4-game-ui-layout
```
contract, and collects stable frames at baseline, narrowest, widest viewports, breakpoint adjacent values, same width different heights, portrait/landscape, default/large font sizes, default/longest copy and zero/non-zero safe areas according to Responsive Visual Validation. Each viewport records viewportRect, canvasRect, logical dimensions, gaps on four sides, background coverage, safe area, key UI boundaries, CSS/physical scaling and changes before/after resize; verify dynamic text, member show/hide, key action states, scroll ownership and resize/orientation rearrangement; device DPR is dynamically read at runtime and capped at 1.5, report records the actual valid value and verifies it is within (0,1.5]. Formal reports must be bound to candidate SHA, scene/state, layout contract version and visual baseline version; render reconstruction is also bound to target SHA, and fidelity cases reference this report. Missing complete viewport can only be marked as "unverified"; ordinary tests assert relationship invariants, and Golden only verifies precise visuals in frozen viewport, valid DPR, language, state and stable frames.
V5 checks structured integration, gameplay ownership, low-fidelity zero references, performance peaks and functional contracts, and submits dynamic evidence for F3. Revised candidates re-run affected F0-F3; only A4-A6 re-run F4. Module boundary changes enter grilling only when there are substantial trade-offs.
After shared foundation is completed, verify representative Vite/Phaser startup, Boot/first scene, resource loading and applicable plugins; after key gameplay flow is completed, smoke test core loop and failure recovery on main target platforms; only final integration/G3/release executes full channel/platform matrix. Partial task evidence cannot claim full platform pass, and cannot automatically initiate real-device acceptance.

Mark as "unverified" when there is no reproducible evidence. V1/V2 condition gates must be bound to the current candidate; silence, continued work or old decisions cannot replace the required

USER_DECISION

, and user selection records cannot skip the deterministic machine F2 of V2a/V2b. FIT only proves proportional no overflow, not full screen or responsive rearrangement; it is prohibited to falsely pass with build success, element existence or no console errors.

Reusable automation scripts are located at

scripts/responsive-visual-validation.mjs

, and pure calculation tests are at

scripts/responsive-visual-validation.test.mjs

. The script dynamically imports Playwright via Node ESM; at runtime, it reads DPR from the device and caps it at

dynamic-capped-1.5

, adjusts the viewport via

setViewportSize

on the same page while retaining resize semantics; valid declarations must be

(0,1.5]

, original

deviceScaleFactor

can be greater than 1.5 but is uniformly parsed and recorded as 1.5, and zero, negative numbers, non-finite numbers and strings fail before verification. The report outputs full page screenshots, runtime measurement matrix and structured JSON.

phaser4-game-qa-performance

NPX Install

Tags

SKILL.md Content (Chinese)

Phaser 4 Testing and Performance

Global Control Access