Claude Cursor GitHub Copilot opencode Skill

quality-performance

Speed, accessibility, and testing bars every delivered app must clear — Core Web Vitals, WCAG 2.2 AA, TDD real-browser E2E, and the build-fail gates.

LLM Mart · 0 points · 1 views 0 listing impressions 0 install-command copies
Download heymegabyte-agent-skills-25-quality-performance-75f4661.zip · 1 KB

Install

skills CLI npx skills add https://github.com/heymegabyte/agent-skills/tree/main/25-quality-performance
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install heymegabyte-agent-skills@llmmart
Git git clone https://github.com/heymegabyte/agent-skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole heymegabyte/agent-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Quality & Performance — every app must clear these bars

Core Web Vitals (per route)

  • LCP ≤ 2.0s · CLS ≤ 0.05 · INP ≤ 100ms (cinematic target; ≤ 200ms hard ceiling). Per-route asset budget enforced.
  • Lighthouse: Performance ≥ 75 · Accessibility ≥ 95, run in CI (Lighthouse CI).

Accessibility (WCAG 2.2 AA)

  • axe-core 0 violations at all 6 breakpoints (375 · 390 · 768 · 1024 · 1280 · 1920). ADA Title II ready.
  • Everything keyboard-operable; visible focus; honors prefers-reduced-motion. (Contrast/target-size specifics live in visual-experience.)

Testing (TDD-first, real flows)

  • Failing Playwright E2E spec BEFORE implementation — watch RED → GREEN. Bug fix = failing regression test FIRST. No feature without ≥1 E2E; no fix without ≥1 regression.
  • E2E runs against the PROD URL in a real browser: goto('/') then navigate by clicks/keyboard only (no page.goto after load). Deterministic (locator waits, no sleeps), parallel-safe, hermetic.
  • Every clickable element · form field · nav link · API endpoint · modal · keyboard shortcut · error/empty/loading state has ≥1 E2E asserting it against prod.
  • Coverage matrix: fullyParallel × 6 viewports × 3 browsers, sharded. Units via Vitest.
  • Simulate realistic full user journeys (homepage → discover → convert) via Cloudflare Browser Rendering or Browserbase (whichever is configured). Drive HIGH % code coverage from these end-to-end journey runs, not just units.

Build-fail gates (any one = fail)

  • Console errors · CSP violations · 4xx/5xx · axe violations · broken hyperlinks — in ANY test — fail the build.
  • AI-vision QA ≥ 8/10 on every new/changed route (random snapshot sampling per run). Visual regression via Percy / pixelmatch.

Mocked ≠ green

  • A mocked/stubbed render is NOT a pass. "Deploy succeeded" ≠ "works" — only a real-browser prod run with 0 console errors and asserted content counts as green.
Files (agent-skills)
  • SKILL.md 2.2 KB
    ---
    name: "quality-performance"
    description: "Speed, accessibility, and testing bars every delivered app must clear — Core Web Vitals, WCAG 2.2 AA, TDD real-browser E2E, and the build-fail gates."
    triggers:
      - "performance"
      - "accessibility"
      - "a11y"
      - "test"
      - "quality"
    priority: 2
    pack: "testing"
    stage: stable
    ---
    
    # Quality & Performance — every app must clear these bars
    
    ## Core Web Vitals (per route)
    
    - LCP ≤ 2.0s · CLS ≤ 0.05 · INP ≤ 100ms (cinematic target; ≤ 200ms hard ceiling). Per-route asset budget enforced.
    - Lighthouse: Performance ≥ 75 · Accessibility ≥ 95, run in CI (Lighthouse CI).
    
    ## Accessibility (WCAG 2.2 AA)
    
    - axe-core 0 violations at all 6 breakpoints (375 · 390 · 768 · 1024 · 1280 · 1920). ADA Title II ready.
    - Everything keyboard-operable; visible focus; honors `prefers-reduced-motion`. (Contrast/target-size specifics live in `visual-experience`.)
    
    ## Testing (TDD-first, real flows)
    
    - Failing Playwright E2E spec BEFORE implementation — watch RED → GREEN. Bug fix = failing regression test FIRST. No feature without ≥1 E2E; no fix without ≥1 regression.
    - E2E runs against the PROD URL in a real browser: `goto('/')` then navigate by clicks/keyboard only (no `page.goto` after load). Deterministic (locator waits, no sleeps), parallel-safe, hermetic.
    - Every clickable element · form field · nav link · API endpoint · modal · keyboard shortcut · error/empty/loading state has ≥1 E2E asserting it against prod.
    - Coverage matrix: `fullyParallel` × 6 viewports × 3 browsers, sharded. Units via Vitest.
    - Simulate realistic full user journeys (homepage → discover → convert) via **Cloudflare Browser Rendering or Browserbase** (whichever is configured). Drive HIGH % code coverage from these end-to-end journey runs, not just units.
    
    ## Build-fail gates (any one = fail)
    
    - Console errors · CSP violations · 4xx/5xx · axe violations · broken hyperlinks — in ANY test — fail the build.
    - AI-vision QA ≥ 8/10 on every new/changed route (random snapshot sampling per run). Visual regression via Percy / pixelmatch.
    
    ## Mocked ≠ green
    
    - A mocked/stubbed render is NOT a pass. "Deploy succeeded" ≠ "works" — only a real-browser prod run with 0 console errors and asserted content counts as green.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.