Deep INP (Interaction to Next Paint) testing: decomposes input delay, processing duration, and presentation delay via the web-vitals/attribution build, asserts per-interaction INP budgets in Playwright using PerformanceObserver plus the web-vitals visibilitychange flush, and identifies long tasks blocking the main thread. Use when a page feels unresponsive while LCP and CLS are green, or to gate key interactions (form submit, modal open, route change) under an INP budget in CI. Covers interactions only - for page-load Web Vitals gating use lighthouse-perf; for service-worker cache-strategy latency use the qa-pwa plugin's service-worker skills.

Authors XCUIest UI tests for iOS / iPadOS / tvOS and macOS desktop apps - uses the three-class XCUIApplication / XCUIElement / XCUIElementQuery pattern, sets accessibility identifiers on production code, runs via `xcodebuild test` with destination, parses the `xcresult` bundle. The macOS desktop delta (platform=macOS destination flags, TCC privacy-permission resets, per-device performance baselines) is in references/macos.md. Use when an iOS or macOS app needs UI tests in Apple's first-party framework (no external runtime; native to Xcode).

Assigns an ML model to a low, medium, or high risk tier from what its predictions decide about people, then derives the fairness and explainability evidence that tier must produce: group metrics per declared sensitive feature, intersectional breakdowns with per-cell counts, vulnerability scan categories, a drift monitoring plan, and per-prediction explanation logs. Supplies conventional demographic parity difference bands, a per-vulnerability-category blocking table, evidence rules marking a bundle incomplete or self-contradicting, and a fairness gating workflow that walks a candidate's model card + evidence bundle to a promote / needs-work / block verdict with refuse rules; a reference covers producing the explanation records with Alibi Explain. Use when a model release candidate is up for promotion and someone must decide which fairness artifacts are mandatory, when a declared risk tier's evidence bundle must be checked against what the tier demands, or when the evidence review must gate the promotion.

Authors k6 JavaScript load-test scripts (VU loops + checks + sleeps), configures the `options` block with `stages` (ramp-up patterns) and `thresholds` (p(95) latency, error rate), runs via `k6 run script.js` or `--vus / --duration` ad-hoc flags, and uses thresholds as the CI pass/fail signal. Includes a latency-percentile interpretation reference: tail ratio (p99/p50), bimodal-distribution detection, coordinated omission and why naive p99 is optimistic, and constant-vus vs constant-arrival-rate executors. Use when the project ships HTTP / WebSocket / gRPC load tests and the team wants developer-friendly JavaScript authoring, or when a k6 threshold passes but the system still feels slow.

Reads CPU flame-graph output from py-spy (Python), async-profiler (JVM), Go pprof, or Node.js `perf_hooks` / clinic.js: identifies the hot path (top sample-time frames), classifies the bottleneck (CPU-bound vs lock contention vs allocator pressure), and proposes the next investigation step. Use when a perf regression is bisected to a commit but the hot path inside it is unclear; for tail-latency percentiles use the latency-percentiles reference in k6-load-testing, and for a slow SQL hot path use db-query-plan-analyzer.

Wires Langfuse tracing into LLM apps for production observability, monitoring, telemetry, and offline eval - instruments via `@observe` (Python) / `startActiveObservation` (TS) decorators that auto-capture inputs / outputs / timings / errors per generation; exposes `langfuse.update_current_span()` for metadata + cost / latency annotation; supports trace-bound scoring for eval datasets and prompt-as-code management. Use when the user needs to monitor, log, trace, or debug LLM API calls in production beyond pre-deploy eval, wants to add LLM observability tooling to an existing app, or wants to ship traces from production to an eval dataset for offline regression testing.

Reference catalog of GDPR-aligned test patterns - data-subject-rights workflows (Art. 15 access, Art. 16 rectification, Art. 17 erasure / "right to be forgotten", Art. 18 restriction, Art. 20 portability, Art. 21 objection); consent recording + revocation per Art. 7; data-residency assertions per Art. 44 - 50 international transfers; breach-notification timing tests per Art. 33 (72 hours); data-minimization assertions in fixtures per Art. 5(1)(c). The California analogue - CCPA/CPRA patterns by Cal. Civ. Code section, including Global Privacy Control (GPC) opt-out, right-to-know, deletion, right-to-correct, and sensitive-PI limits - lives in references/ccpa.md. Use when authoring GDPR- or CCPA/CPRA-readiness tests for any product processing EU or California personal data.

Author and operate Selenium Grid 4 - self-hosted distributed WebDriver. Covers the six-component architecture (Router / Distributor / Session Map / Event Bus / New Session Queue / Node), standalone vs hub-and-node modes, the Docker-image stack (selenium/standalone-chrome, selenium/hub, selenium/node-chrome), node registration, session-queue tuning, and observability. Use for self-hosted cross-browser testing when data residency or cost-control require an on-prem grid. This is the self-hosted execution RUNNER - for the zero-infra alternative use playwright-testing browser projects (bundled engines); for managed cloud grids use cloud-grid-e2e (BrowserStack / Sauce Labs / LambdaTest); to decide WHICH browsers and tiers to run use browser-matrix-strategy-reference.

Orchestrates WireMock fault stubs (HTTP-level fault: 500s, malformed JSON, slow responses) with Toxiproxy (TCP-level: latency, packet loss, reset) into a single resilience test scenario - the test starts both, applies fault per scenario, runs the SUT against the impaired endpoints, verifies the SUT's resilience patterns. Use when one test must reproduce a combined network + HTTP failure - a cross-layer failure mode from an incident postmortem that neither pure HTTP fault stubs nor pure TCP chaos can cover alone, because most real failures span both layers.

Run protocol and run workflow for a chaos experiment that has already been designed: the four pre-flight gates (non-production target, measured healthy baseline, live observability, a rollback that has actually been exercised), how to pick a conservative blast-radius bound, the sampling cadence and abort criteria fixed in writing before injection, the per-runner inject and abort commands (Chaos Mesh / Litmus / Gremlin / Toxiproxy), the refuse-to-start rules (no blast-radius bound, production context, degraded baseline, offline observability, unexercised rollback), and the recovery-validation step with its tolerance and timeout. Owns execution safety only, not experiment design: the steady-state hypothesis, the fault to inject, and the experiment file come from chaos-experiment-author. Use when an experiment definition exists and a fault is about to be injected into a running system, and the go/no-go gates, abort thresholds, and recovery check still need to be agreed and written down before the fault starts.

Builds the full manual-accessibility artifact surface: step-by-step screen-reader test scripts for NVDA (Windows), JAWS (Windows), VoiceOver (macOS / iOS), or TalkBack (Android) with per-step keystroke + expected announcement; per-archetype WCAG 2.2 checklists (references/wcag-checklist.md); per-widget keystroke matrices pairing expected NVDA and VoiceOver announcements with the WCAG SC each row verifies (references/widget-matrix.md); and a guided NVDA / VoiceOver session protocol that merges script + checklist into a signed pass/fail session report. Use when authoring an accessibility-acceptance test, checklist, or widget matrix the team will run before sign-off, when scripting a manual a11y audit, OR when walking a tester through a guided screen-reader session.