Token导航 LogoToken导航

Agent Skill · 研究检索

astro-performance(天文表演)

astro-performance 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

astro-performance 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 59,000。

总安装 396周安装 17GitHub Stars 2下载 139核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:astro-performance(天文表演)
来源仓库:https://github.com/soborbo/claudeskills
仓库路径:skills/astro-performance
安装命令:
npx skills add https://github.com/soborbo/claudeskills --skill astro-performance
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/soborbo/claudeskills --skill astro-performance

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

用于提升网站性能和核心 Web Vitals 指标。提供 Lighthouse 评分优化方案和关键渲染路径控制。需遵循三跳规则,适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。避免字体加载和额外 CSS 增加 LCP 延迟。

astro-performance 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/astro-performance

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/soborbo/claudeskills --skill astro-performance;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

astro-performance 怎么安装?

在终端执行 npx skills add https://github.com/soborbo/claudeskills --skill astro-performance。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

astro-performance 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

astro-performance 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

astro-performance 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

astro-performance 用的人多吗?

公开数据是累计安装 396 次,近一周安装 17 次,下载 139 次,GitHub 2 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

astro-performance 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
astro-performance

天文表演 · 本页

3962需要联网未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Astro Performance Skill

Purpose

Achieve 90+ Lighthouse mobile scores and pass Core Web Vitals on every page. Direct impact on SEO rankings, crawl priority, and conversion rates.

Core Web Vitals Targets

MetricGoodNeeds ImprovementPoor
LCP (Largest Contentful Paint)≤2.5s2.5–4s>4s
INP (Interaction to Next Paint)≤200ms200–500ms>500ms
CLS (Cumulative Layout Shift)≤0.10.1–0.25>0.25

Critical Path Rule

The critical rendering path must be max 3 hops: HTML → CSS → LCP image.

Anything that adds a 4th hop (font in CSS chain, extra CSS file, unpreloaded image) adds 150–500ms to LCP on mobile. This is the #1 cause of poor mobile scores.

GOOD:  HTML (150ms) → Layout.css (150ms) → Hero image (preloaded, parallel)
       Total: ~300ms to FCP, ~500ms to LCP

BAD:   HTML (150ms) → Layout.css (150ms) → italic font (350ms) → Hero image (discovered late)
       Total: ~650ms to FCP, ~2000ms+ to LCP

Core Rules

1. LCP Preloading (Biggest Impact)

Every page MUST pass its hero image to BaseLayout for preloading:

<BaseLayout
  preloadImage="/img/hero-480w.avif"
  title="Page Title"
>

BaseLayout renders: <link rel="preload" as="image" href="..." type="image/avif" fetchpriority="high">

  • The preloadImage prop auto-detects MIME type from extension
  • Each page passes its OWN hero image — never hardcode a single image for all pages
  • Only ONE fetchpriority="high" per page (the preload + the <img> tag)

2. Font Strategy

Primary fonts (body, headings): font-display: swap + preload in <head> Non-critical variants (italic, display, decorative): font-display: optional + lazy CSS

<!-- Primary: preload + swap (in <head>) — font from siteConfig.fonts -->
<link rel="preload" as="font" type="font/woff2" href="/fonts/body-font.woff2" crossorigin>

<!-- Non-critical: lazy load (NOT in <head> as blocking) -->
<link rel="stylesheet" href="/fonts/body-font-italic.css" media="print" onload="this.media='all'">
<noscript><link rel="stylesheet" href="/fonts/body-font-italic.css"></noscript>

Why: A font in the main CSS adds it to the critical path chain (HTML → CSS → font = +300–500ms). Moving non-critical fonts to lazy CSS removes them from the chain entirely.

Never preload non-critical fonts — that defeats the lazy loading.

3. Images

Use the <Picture> component from astro-images skill with pattern-based srcset. See images reference for details.

Key rules:

  • Three formats always: AVIF → WebP → JPG (never PNG fallback for photos)
  • 480w variant in every pattern
  • loading="eager" on hero and above-fold only
  • loading="lazy" on everything below-fold
  • Never loading="lazy" on hero. Never loading="eager" on below-fold.
  • Explicit width/height on ALL images including SVGs
  • width/height must match delivered dimensions, not original source

4. Third-Party Scripts

Defer everything that isn't essential for first render.

<!-- GTM ID comes from siteConfig.tracking.gtmId — only deferGtmMs is a prop -->
<BaseLayout title="Page Title" deferGtmMs={2000}>

<!-- FB Pixel: always setTimeout if loaded directly -->
<!-- Tag Gateway: managed in Cloudflare Dashboard, NOT in code -->

See third-party scripts reference for Cloudflare Tag Gateway details.

5. CSS Strategy

  • ONE main CSS file (Layout.css) — this is the only render-blocking CSS allowed
  • Component CSS that's only used below-fold: use <style is:inline> to move it into the HTML body (not a blocking <link> in <head>)
  • Never add extra <link rel="stylesheet"> in <head> unless absolutely needed above-fold
  • Below-fold component CSS total impact: monitor, should not add >2KB inline

6. CLS Prevention

  • Every <img> and <iframe> needs width + height attributes
  • SVG <img> tags included — read dimensions from SVG viewBox
  • Use aspect-ratio CSS as backup
  • Reserve space for dynamic content with min-height
  • See CLS reference

7. Bundle Size

Asset TypeBudget
Total JS (own code)<50KB gzipped
Total JS (with 3rd party)<100KB gzipped
Total CSS<50KB gzipped
Hero image<200KB
Any single image<100KB
OG images<150KB each (JPG q80)

8. Caching

  • Hashed assets (/_astro/*): Cache-Control: public, max-age=31536000, immutable
  • HTML: Cache-Control: public, max-age=0, must-revalidate
  • Fonts: Cache-Control: public, max-age=31536000, immutable
  • See caching reference

Lighthouse Score Variability

Lighthouse mobile scores fluctuate ±10–15 points between runs. This is normal.

The slow 4G emulation is non-deterministic — the same page can score 65 on one run and 92 on the next. Same CSS file might take 150ms or 330ms to "load" in the emulation.

Best practice:

  • Run 3–5 times and take the median score
  • Use lighthouse --preset=perf locally for consistent results
  • Google uses real user data (CrUX) for ranking, not single Lighthouse runs
  • Don't chase single-digit improvements after 90+
# Local batch testing
for i in 1 2 3 4 5; do
  lighthouse https://example.com --only-categories=performance \
    --preset=perf --form-factor=mobile --output=json \
    --output-path="./lh-run-$i.json" --chrome-flags="--headless"
  score=$(cat "lh-run-$i.json" | node -e "process.stdin.on('data',d=>console.log(Math.round(JSON.parse(d).categories.performance.score*100)))")
  echo "Run $i: $score"
done

Subpage Performance Checklist

Don't only test the homepage. Different page types have different performance profiles.

Test at minimum:

  • Homepage /
  • Longest service page (most content)
  • An area/location page
  • Reviews page (many DOM elements)
  • Calculator page (JS-heavy)

Common subpage-specific issues:

  • Large inline JSON-LD schema in <head> → move to </body> or minimize
  • Many review cards rendered at once → large DOM, slow layout
  • Missing preloadImage prop → hero image discovered late
  • Extra <style is:inline> blocks from components → HTML size bloat

Integration with Other Skills

SkillHow it connects
astro-imagesUse <Picture> component with patterns. LCP image = lcp prop.
design-tokensColor contrast (WCAG AA 4.5:1) — poor contrast = a11y failure, not perf, but Lighthouse reports both
schema-entity-graphSitemap <lastmod> must sync with dateModified in schema — not a perf issue but often fixed in same pass
deploymentPre-deploy checks, Cloudflare Workers config, output: 'static' for build-time image processing

Boilerplate

On first use in a project, copy the BaseLayout:

cp assets/boilerplate/layouts/BaseLayout.astro → src/layouts/BaseLayout.astro

Skip if the project already has it. The BaseLayout handles LCP preloading, GTM deferral, Schema.org rendering, and meta tags.

References

Core Web Vitals

Assets & Resources

  • Bundle Size — Analysis, tree shaking, dynamic imports
  • Fonts — Swap vs optional, lazy CSS for non-critical, subsetting
  • Images — Pattern-based srcset, format priority, SVG rules

Infrastructure

  • Third-Party Scripts — GTM defer, Tag Gateway, FB Pixel, facade pattern
  • Caching — Cloudflare headers, cache control
  • Testing — Lighthouse CLI, batch runs, real user monitoring

Forbidden

  • Extra <link rel="stylesheet"> in <body> (use <style is:inline> instead for below-fold component CSS)
  • Extra <link rel="stylesheet"> in <head> for below-fold components
  • Synchronous third-party scripts in <head> (defer or use deferGtmMs)
  • PNG fallback for photo images (use JPG)
  • Unoptimized images / missing AVIF+WebP variants
  • font-display: swap on non-critical font variants (use optional + lazy CSS)
  • Non-critical font variants loaded in render-blocking CSS
  • font-display: block (blocks rendering up to 3s)
  • Preloading non-critical fonts (defeats lazy loading)
  • loading="lazy" on hero images
  • loading="eager" on below-fold images
  • Missing width/height on any <img> or <iframe> (including SVGs)
  • width/height set to original source dimensions instead of delivered size
  • Layout shifts from dynamic content without reserved space
  • Main thread blocking >50ms without chunking
  • More than ONE fetchpriority="high" per page
  • Hardcoded preload URL in Layout (use preloadImage prop per page)
  • Modifying Cloudflare Tag Gateway scripts (/ry2s/) in HTML

Definition of Done

  • Lighthouse mobile ≥90 on homepage (median of 3 runs)
  • Lighthouse mobile ≥85 on worst subpage (median of 3 runs)
  • LCP ≤2.5s (homepage) / ≤3.5s (subpages)
  • CLS ≤0.1 on all pages
  • INP ≤200ms on calculator/interactive pages
  • Critical path: max 3 hops (HTML → CSS → LCP image)
  • No font in critical path chain (italic/display = lazy CSS)
  • Total own JS <50KB gzipped
  • Every page has preloadImage prop with correct hero
  • Hero image preloaded, loading="eager", fetchpriority="high"
  • All below-fold images: loading="lazy"
  • All <img> and <iframe> have width + height (including SVGs)
  • Fonts self-hosted with correct display strategy
  • Third-party scripts deferred (GTM: deferGtmMs, FB: setTimeout)
  • OG images generated from hero (5 variants, JPG)
  • Tested on ≥3 different page types, not just homepage

平台分布

Codex

34.59%
按下载量换算48

Claude

30.24%
按下载量换算42

Cursor

19.25%
按下载量换算27

Gemini CLI

9.64%
按下载量换算13

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills