AGENT SKILL

frontend-lighthouse

Add a portable Lighthouse CI gate for production frontend builds with Core Web Vitals budgets, category floors, median runs, and CI artifacts.

License 许可
MIT
最近更新
2026-10-22
Stars
★ 1
Forks
0

01 前言 Front Matter

name
frontend-lighthouse
risk
safe
tags
[ "frontend", "lighthouse", "performance", "core-web-vitals", "ci" ]
tools
[ "lighthouse", "node", "github-actions" ]
author
stareezy-1
source
community
license
MIT
category
frontend
date_added
2026-06-29
description
Add a portable Lighthouse CI gate for production frontend builds with Core Web Vitals budgets, category floors, median runs, and CI artifacts.
source_repo
stareezy-1/frontend-architecture-skill
source_type
community
license_source
https://github.com/stareezy-1/frontend-architecture-skill/blob/main/LICENSE

03 SKILL.md 正文

Frontend Lighthouse (portable performance gate)

Portable skill — readable by Claude Code, OpenCode, Codex, Cursor, Windsurf, and others. This skill describes a CI performance gate — a Lighthouse CI config plus a workflow — not a component library or a visual style. It pairs with the frontend-seo and frontend-architecture skills: SEO writes the metadata, Lighthouse proves it ships fast.

The goal: every pull request is blocked unless the production build meets explicit Core Web Vitals budgets and category score floors. Budgets live in one lighthouserc.cjs, runs are median-of-N so the gate doesn't flake, and the same config runs locally and in CI.

When to Use This Skill

  • Use when adding a Lighthouse CI performance gate to a web app.
  • Use when setting Core Web Vitals budgets for LCP, CLS, and TBT as the lab proxy for INP.
  • Use when configuring category score floors for performance, SEO, accessibility, and best practices.
  • Use when debugging flaky Lighthouse runs or making reports visible as CI artifacts.

0. The five core ideas

  1. One config, one source of truth. All budgets and assertions live in a single lighthouserc.cjs. Named constants for each budget — no magic numbers buried in assertion objects.
  2. Gate the production build, never dev. Lighthouse runs against build + start (the real, optimized output). Dev-server numbers are meaningless for a budget.
  3. Median-of-N kills flakiness. Run 3+ times and assert on the median run, so per-run jitter (cold caches, CI noise) never red-flags a healthy build.
  4. Budgets encode Google's "good" thresholds. LCP ≤ 2500 ms, INP ≤ 200 ms (gated via the TBT lab proxy), CLS ≤ 0.1 — the values that earn green scores, not "needs improvement".
  5. Blocking in CI, visible as artifacts. A GitHub Action runs the gate on every PR touching the app and uploads the HTML/JSON reports so failures are debuggable.

1. Files this skill adds

apps/web/                          (or your app root)
├── lighthouserc.cjs               ← the gate: budgets + assertions + collect settings
├── package.json                   ← "lhci": "lhci autorun --config=./lighthouserc.cjs"
└── .github/workflows/lighthouse.yml  ← PR-blocking CI job (build → start → lhci → upload)

Plus a dev dependency: @lhci/cli.

pnpm add -D @lhci/cli        # or npm i -D / yarn add -D

2. The config (lighthouserc.cjs)

.cjs (CommonJS) so it loads without ESM/TS transpilation. Every budget is a named constant with a comment explaining the threshold — never a bare number inside an assertion.

/**
 * Lighthouse CI configuration — Core Web Vitals budgets for the marketing surface.
 *
 * Enforces Google's mobile "good" CWV thresholds:
 *   - Largest Contentful Paint (LCP) ≤ 2500 ms
 *   - Cumulative Layout Shift (CLS)  ≤ 0.1
 *   - Interaction to Next Paint (INP) ≤ 200 ms
 *
 * INP is a *field* metric with no direct lab audit, so in the lab we gate on
 * Total Blocking Time (TBT) — Lighthouse's recommended lab proxy — at the same
 * budget, and assert the experimental INP audit directly as a warning where the
 * build exposes it.
 *
 * Collection runs against the *production* server (build + start) on Lighthouse's
 * default mobile (Moto G4 / slow 4G) emulation.
 */

/** The fixed port the production server is started on for the audit. */
const PORT = 3100;
const BASE_URL = `http://localhost:${PORT}`;

/** Pages whose budgets are enforced in CI. */
const MARKETING_URLS = [`${BASE_URL}/`];

/**
 * Core Web Vitals budgets on mobile — Google's "good" thresholds.
 * These are the values that earn the best Lighthouse scores.
 */
const LCP_BUDGET_MS = 2500; // good
const INP_BUDGET_MS = 200; // good (TBT lab proxy)
const CLS_BUDGET = 0.1; // good

module.exports = {
  ci: {
    collect: {
      // Build is run separately in CI; here we only serve the production output.
      startServerCommand: `pnpm start --port ${PORT}`,
      startServerReadyPattern: "Ready in", // framework's "server ready" log line
      startServerReadyTimeout: 120000,
      url: MARKETING_URLS,
      // Median of multiple runs keeps the gate stable against per-run jitter.
      numberOfRuns: 3,
      settings: {
        // Default mobile emulation; opt into desktop via env for a second run.
        preset:
          process.env.LHCI_FORM_FACTOR === "desktop" ? "desktop" : undefined,
        // Only gate the categories we care about; skip PWA category noise.
        onlyCategories: [
          "performance",
          "seo",
          "accessibility",
          "best-practices",
        ],
      },
    },
    assert: {
      // Median across runs is the value compared against each budget.
      aggregationMethod: "median-run",
      assertions: {
        // --- Core Web Vitals budgets (the contract) ---------------------
        "largest-contentful-paint": [
          "error",
          { maxNumericValue: LCP_BUDGET_MS },
        ],
        "cumulative-layout-shift": ["error", { maxNumericValue: CLS_BUDGET }],
        "total-blocking-time": ["error", { maxNumericValue: INP_BUDGET_MS }],
        // Direct INP audit where the Lighthouse build exposes it (else ignored).
        "interaction-to-next-paint": [
          "warn",
          { maxNumericValue: INP_BUDGET_MS },
        ],

        // --- Category floors (target top Lighthouse scores) -------------
        "categories:performance": ["error", { minScore: 0.9 }],
        "categories:seo": ["error", { minScore: 0.95 }],
        "categories:accessibility": ["error", { minScore: 0.95 }],
        "categories:best-practices": ["error", { minScore: 0.9 }],
      },
    },
    upload: {
      // Keep reports in the CI run's filesystem; no external LHCI server.
      target: "filesystem",
      outputDir: "./.lighthouseci",
    },
  },
};

Hard rules:

  • Every budget is a named constant with a unit in its name (LCP_BUDGET_MS) and a comment.
  • aggregationMethod: "median-run" is non-negotiable — single-run gates flake constantly.
  • numberOfRuns ≥ 3 (odd numbers give a clean median).
  • Assert on TBT for INP in the lab; treat the experimental interaction-to-next-paint audit as a warn, not an error (it isn't present in every Lighthouse build).
  • Keep onlyCategories to exactly what you gate — fewer audits, faster, less noise.

3. Choosing budget severity and thresholds

Audit / category Severity Threshold Why
largest-contentful-paint error ≤ 2500 ms Google "good" LCP
cumulative-layout-shift error ≤ 0.1 Google "good" CLS
total-blocking-time error ≤ 200 ms INP lab proxy
interaction-to-next-paint warn ≤ 200 ms not in all builds; don't hard-fail on a missing audit
categories:performance error ≥ 0.9 top (green) band
categories:seo error ≥ 0.95 SEO is cheap to keep perfect
categories:accessibility error ≥ 0.95 a11y regressions must block
categories:best-practices error ≥ 0.9 green band

Use error for contracts that must hold and warn for audits that are environment-dependent or aspirational. Start strict and only loosen with a recorded reason — a budget you keep raising to make CI pass is a budget that no longer protects anything.


4. The npm script

// package.json
{
  "scripts": {
    "lhci": "lhci autorun --config=./lighthouserc.cjs"
  }
}

lhci autorun runs collect → assert → upload in sequence. Run it locally before pushing to reproduce exactly what CI does:

pnpm build && pnpm lhci
# desktop form factor:
LHCI_FORM_FACTOR=desktop pnpm build && LHCI_FORM_FACTOR=desktop pnpm lhci

5. The GitHub Actions workflow

Runs on PRs that touch the app or the workflow itself. Builds the production output, runs the gate, and always uploads the reports (even on failure) so a red check is debuggable.

name: Lighthouse CWV

on:
  pull_request:
    branches: [main]
    paths:
      - "apps/web/**"
      - ".github/workflows/lighthouse.yml"

permissions:
  contents: read

jobs:
  lighthouse:
    name: Lighthouse CWV (marketing pages)
    runs-on: ubuntu-latest
    defaults:
      run:
        working-directory: apps/web
    steps:
      - uses: actions/checkout@v4

      - name: Setup pnpm
        uses: pnpm/action-setup@v4 # version comes from root package.json packageManager

      - name: Setup Node
        uses: actions/setup-node@v4
        with:
          node-version: 22
          cache: pnpm

      - name: Install dependencies
        working-directory: .
        run: pnpm install --frozen-lockfile

      - name: Build web app
        run: pnpm build

      # build + start the production server, run Lighthouse on mobile emulation,
      # fail the job if any budget in lighthouserc.cjs is exceeded.
      - name: Run Lighthouse CI
        run: pnpm lhci

      - name: Upload Lighthouse reports
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: lighthouse-reports
          path: apps/web/.lighthouseci
          if-no-files-found: ignore

Hard rules:

  • Trigger on the app path and the workflow file so config changes are self-testing.
  • if: always() on the upload step — you need the report most when the gate fails.
  • Gate on the production build (pnpm build then the start server in collect).
  • Match the CI Node/pnpm versions to the repo's pinned versions to avoid lockfile drift.

6. Framework adapters

The config is framework-neutral except startServerCommand and startServerReadyPattern.

Framework startServerCommand startServerReadyPattern
Next.js pnpm start --port 3100 (after next build) "Ready in"
Remix pnpm start (serve the built app) server's listening log line
Astro node ./dist/server/entry.mjs (SSR) or npx serve dist (static) the adapter's ready line / serve's URL line
SvelteKit node build (node adapter) "Listening on"
Vite SPA npx vite preview --port 3100 "Local:"

For purely static output you can skip the server and point collect.staticDistDir at the build folder instead of startServerCommand — Lighthouse serves it internally.


7. Debugging failing or flaky runs

  • Flaky LCP/TBT → raise numberOfRuns (5), confirm median-run, and make sure nothing else is competing for CPU on the runner.
  • interaction-to-next-paint errors → it should be warn, not error; the audit is missing in some Lighthouse versions.
  • "server not ready" timeout → fix startServerReadyPattern to match the framework's actual ready log, and raise startServerReadyTimeout.
  • Real regressions → open the uploaded report artifact, read the failed audit's "Opportunities"/"Diagnostics", fix the cause (oversized image, render-blocking JS, layout shift from unsized media) — don't just bump the budget.
  • Desktop vs mobile divergence → run both form factors; mobile is the stricter gate and should be the default.

8. Conventions checklist (enforce in review)

  • [ ] All budgets are named constants with units and comments — no magic numbers in assertions.
  • [ ] Gate runs against the production build, never the dev server.
  • [ ] aggregationMethod: "median-run" with numberOfRuns ≥ 3.
  • [ ] CWV budgets at Google "good" thresholds (LCP ≤ 2500, TBT ≤ 200, CLS ≤ 0.1).
  • [ ] INP gated via TBT (error); experimental INP audit is warn.
  • [ ] Category floors set as error (perf ≥ 0.9, SEO/a11y ≥ 0.95, best-practices ≥ 0.9).
  • [ ] onlyCategories lists exactly the gated categories.
  • [ ] CI triggers on the app path and the workflow file; reports upload with if: always().
  • [ ] Local pnpm lhci reproduces the CI run.
  • [ ] Budgets are tightened over time, loosened only with a recorded reason.

9. How to apply this skill

Adding the gate to a project: install @lhci/cli, drop in lighthouserc.cjs with your URLs and startServerCommand, add the lhci script, and add the workflow. Run pnpm build && pnpm lhci locally to confirm it passes before opening a PR.

Adding a page to the gate: append its URL to MARKETING_URLS (or a second URL array). Each URL is audited independently against the same budgets.

Tuning budgets: change the named constant, not the assertion. Record why in the comment. Prefer fixing the regression over raising the budget.

Reviewing performance: run the checklist in §8. The highest-value catches are a gate that runs against the dev server (meaningless numbers) and single-run assertions (chronic flakiness).


Publishing / installing this skill

This skill follows the Anthropic SKILL.md format and is portable across agents.

  1. Keep it under skills/frontend-lighthouse/SKILL.md in a public GitHub repo.
  2. Keep the frontmatter name and high-signal description — discovery indexes match against it.
  3. Install with: npx skills add <org>/<repo> --skill "frontend-lighthouse".
  4. Non-SKILL.md agents can be pointed here from AGENTS.md / CLAUDE.md; Kiro can mirror it as a steering file.

Limitations

  • Lighthouse CI is a lab signal and does not replace field monitoring from real-user metrics.
  • Budgets must be tuned to the actual app route, hosting platform, and device/network assumptions.
  • A passing Lighthouse gate does not prove business-critical flows, visual correctness, or backend availability.

04 同仓库更多 Skills

Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

testing-best-practices

Laravel test design and review. Use when selecting coverage, naming or structuring tests, choosing assertions or test data, isolating dependencies, testing HTTP or security boundaries, improving suite performance, or reviewing test value. Use framework guidance or search-docs for Pest and PHPUnit syntax.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

tailwindcss-development

Always invoke when the user's message includes 'tailwind' in any form. Also invoke for: building responsive grid layouts (multi-column card grids, product grids), flex/grid page structures (dashboards with sidebars, fixed topbars, mobile-toggle navs), styling UI components (cards, tables, navbars, pricing sections, forms, inputs, badges), adding dark mode variants, fixing spacing or typography, and Tailwind v3/v4 work. The core use case: writing or fixing Tailwind utility classes in HTML templates (Blade, JSX, Vue). Skip for backend PHP logic, database queries, API routes, JavaScript with no HTML/CSS component, CSS file audits, build tool configuration, and vanilla CSS.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

laravel-best-practices

Apply this skill whenever writing, reviewing, or refactoring Laravel PHP code. This includes creating or modifying controllers, models, migrations, form requests, policies, jobs, scheduled commands, service classes, and Eloquent queries. Triggers for N+1 and query performance issues, caching strategies, authorization and security patterns, validation, error handling, queue and job configuration, route definitions, and architectural decisions. Also use for Laravel code reviews and refactoring existing Laravel code to follow best practices. Covers any task involving Laravel backend PHP code patterns.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

infer-conventions

Use this skill to analyze how a Laravel application is actually written and record its conventions as shared rules. Trigger when the user wants to detect, infer, document, or standardize project conventions or coding style, set up or grow `.ai/rules`, resolve mixed or conflicting patterns (e.g. "are we using Form Requests or inline validation?"), or onboard agents and teammates to "how we do things here". Covers: a systematic sweep of ~49 Laravel convention dimensions (validation, models, architecture, testing, frontend, database, console), open-ended house-pattern discovery, conflict reporting, and recording rules scoped to the right paths via the Boost `record-rule` MCP tool. Only run this skill when the user explicitly asks for it; never start a sweep as part of another task. Do not use for one-off code review, enforcing formatting a linter already handles, or editing `.ai/rules` files by hand.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

inertia-react-development

Develops Inertia.js v2 React client-side applications. Activates when creating React pages, forms, or navigation; using <Link>, <Form>, useForm, or router; working with deferred props, prefetching, or polling; or when user mentions React with Inertia, React pages, React forms, or React navigation.

★ 1
Hazek-dc
Hazek-dc/PresensiGuru_FaceNet_EMAR

deploying-to-cloud

Deploys and manages Laravel applications on Laravel Cloud using the `cloud` CLI. Use when the user wants to deploy an app, ship to Cloud, create or manage applications, environments, databases, caches, object storage, queues, domains, instances, background processes, secrets, compute, scheduled tasks, or Laravel Cloud infrastructure; or check billing, usage, or spend. Triggers on deployment, shipping, Cloud management, environment setup, database provisioning, billing or usage queries, the `cloud` CLI, and Laravel Cloud deployment troubleshooting.

★ 1