菜单
Audit any website for Agent Experience Optimization (AEO). Checks llms.txt, agent views, schema.org, sitemap, agent-card.json, pricing.md, NLWeb feeds, semantic HTML, visual labeling, and cross-platform consistency. Provides applicability judgment and explicit fix plans.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
use when asked to roast, audit, score, shame, or clean up a local developer environment — messy project folders, home/downloads/desktop clutter, abandoned projects, poor git hygiene, AI slop, leaked-secret risk, unused tooling, dependency graveyards, or hackathon demos — on macOS, Linux, or Windows, without uploading private files.
USE THIS SKILL whenever the user wants to: refactor a codebase, clean up tech debt, improve code consistency, reduce complexity, make a codebase more maintainable, or when the user mentions entropy, code quality, code health, or asks to clean up or simplify a module, directory, or entire repo. Especially valuable for AI-first and vibe coding teams where code accumulates disorder quickly.
Generate channel-consistent titles, descriptions, and tags for YouTube videos. Triggers on: "video description", "YouTube title", "content ops", "channel style".
USE THIS SKILL whenever the user wants to: build a personal knowledge wiki, ingest data into a wiki, compile notes into articles, query a knowledge base, or clean up wiki content. Triggers on 'wiki', 'personal wiki', 'knowledge base', 'compile notes', or 'build wiki'.
USE THIS SKILL whenever the user wants to: audit wiki backlinks, verify link legitimacy, check for false links, or validate cross-references from heavily linked pages. Triggers on 'link audit', 'backlink check', 'verify links', or 'wiki integrity'.
USE THIS SKILL whenever the user wants to: ingest a certificate, contract, employment letter, or other text-based document into a wiki's sources layer and integrate its facts into a wiki page. Triggers on 'add certificate to wiki', 'ingest document into sources', 'create source document from text', or 'add employment info to wiki from document'.
| name | aeo-audit |
| description | Audit any website for Agent Experience Optimization (AEO). Checks llms.txt, agent views, schema.org, sitemap, agent-card.json, pricing.md, NLWeb feeds, semantic HTML, visual labeling, and cross-platform consistency. Provides applicability judgment and explicit fix plans. |
| triggers | ["user says \"audit AEO\"","user says \"agent experience optimization\"","user says \"check agent readable\"","user says \"SEO/AEO audit\"","user says \"make site AI discoverable\"","user says \"agent audit\"","before publishing any marketing site, portfolio, or product site"] |
Audit a website for AI agent discoverability, ingestibility, and consistency. This is the agent-native equivalent of an SEO audit.
Before running checks, classify the site. This determines which items are required, conditional, or irrelevant.
| Site Type | Examples | Key Conditional Items |
|---|---|---|
| Personal portfolio | Creative, developer, photographer | No pricing, no API, no MCP |
| SaaS / Product | App landing, dashboard, tool | pricing.md, API docs, agent-card.json |
| Blog / Publication | Personal blog, newsletter, magazine | No pricing, no MCP, per-section llms.txt per category |
| E-commerce | Store, marketplace | Product schema, pricing, inventory feeds |
| Documentation | API docs, knowledge base | Per-section llms.txt, NLWeb schema, agent-card.json |
| Open-source project | GitHub Pages, library site | MCP server docs, agent-card.json, API endpoints |
Rule: Never recommend pricing.md, agent-card.json, MCP references, or API authentication docs for a site that has no pricing, no API, and no MCP server.
Run every check that is required or conditional-applicable for the site's classification. For each item, record: status (pass / fail / skip / missing), severity (critical / warning / info), and fix (explicit action).
/sitemap.xml)curl -s https://<site>/sitemap.xml | head -n 5<urlset> or <sitemapindex>app/sitemap.ts. Static: use generator script./robots.txt)curl -s https://<site>/robots.txtSitemap: directive, does not blanket-disallow /Sitemap: line, or Disallow: /robots.txt:
User-agent: *
Allow: /
Sitemap: https://<site>/sitemap.xml
curl -I https://<site>/ and inspect Link: headerrel="llms.txt", rel="sitemap", rel="robots" (and agent endpoints if they exist)Link header, or references non-existent endpointsheaders() in next.config.ts (or server config). Only include endpoints that return 200.curl -s https://<site>/.well-known/llms.txt/llms.txt (canonical alias)/llms.txt exists, create redirect or duplicate at /.well-known/llms.txt./llms.txt existscurl -s https://<site>/llms.txt# (top-level heading), contains Overview, Contact, and Machine-Readable Endpoints sectionsapp/llms.txt/route.ts (Next.js) or static file. Follow template in web-agent-discoverability skill./llms-full.txt existscurl -s https://<site>/llms-full.txt | wc -c/docs/llms.txt, /api/llms.txt, etc./docs/, /api/, /developers/, /blog/ sectionscurl -s https://<site>/docs/llms.txt/.well-known/llms.txt (canonical)/llms.txt/.well-known/llms.txt/llms.txt/llms.txt is already well-known enough?mode=agent query param on homepagecurl -s "https://<site>/?mode=agent" | grep -o '<h1>'<h1>, <h2>, <ul>, <a> (not the same marketing HTML)mode=agent and render <AgentView /> with clean semantic HTML./agent JSON endpointcurl -s https://<site>/agent | jq '.'Content-Type: application/json, contains identity, services, content, contact, metaapp/agent/route.ts returning structured JSON./index.md markdown fallbackcurl -s https://<site>/index.md | head -n 3text/markdown, body starts with # app/index.md/route.ts or static file. Content mirrors homepage in Markdown.curl -s https://<site>/ | grep -o '<h1' and curl -s https://<site>/ | wc -c<h1> and >500 chars of meaningful textcurl -s https://<site>/ | grep -o 'application/ld+json' | wc -l and inspect content"@type": "Person" or "@type": "Organization" with name, url, sameAs, descriptionWebPage only), no sameAs<script> in <body> via layout component. Use lib/jsonld.ts pattern."@type": "WebSite" with name, url, description, inLanguagecurl -s https://<site>/projects/<slug> | grep -A 20 'application/ld+json'"@type": "CreativeWork" or "BlogPosting" with name, description, creator, dateCreated"@type": "BreadcrumbList"hasOfferCatalog with Offer itemshasOfferCatalog to Person/Organization schema.curl -s https://<site>/nlweb-schema.json or inspect page for NLWeb markup<title> tag present on every pagecurl -s https://<site>/ | grep -o '<title>[^<]*</title>'metadata.title or <head><title>.<meta name="description"> presentcurl -s https://<site>/ | grep -o 'name="description" content="[^"]*"'metadata.description per page.og:title, og:description, og:type, og:url, og:image)curl -s https://<site>/ | grep -o 'property="og:[a-z]*"'metadata.openGraph in Next.js or <meta property="og:*">.curl -s https://<site>/ | grep -o 'name="twitter:[a-z]*"'twitter:card, twitter:title, twitter:description presentmetadata.twitter in Next.js.<title>, og:title, JSON-LD name, and <meta name="description">content/site.ts).h1 → h2 → h3)curl -s https://<site>/ | grep -o '<h[1-6]' | sort | uniq -ch1, logical progression, no skipped levelsh1s, missing h1, h3 without h2h1.curl -s https://<site>/ | grep -o '<img[^>]*>' | grep -v 'alt=' | wc -l<img> have meaningful alt attributes (or alt="" for decorative)altalt to all next/image or <img> tags.alt or aria-label; galleries have aria-label or role="list"curl -s https://<site>/ | grep -o '<a[^>]*>[^<]*</a>' | grep -E '^<a.*>(click here|read more|link|here)</a>$'agent-card.json existscurl -s https://<site>/agent-card.json | jq '.'name, description, capabilities, endpointspricing.mdcurl -s https://<site>/pricing.md or /pricing/llms.txt or docs mention MCP; /.well-known/mcp/ exists/api/llms.txt, OpenAPI spec, or /agent endpoint describes API/llms.txt and create /api/llms.txt if API exists./llms.txt project list with actual site projects; compare /agent JSON with homepageagents.md in project root documenting which files are auto-synced and which need manual updates. See agents.md pattern in portfolio project.After completing checks, produce a structured report:
## AEO Audit Report — {site-name}
### Summary
- Checks run: {N}
- Passed: {N}
- Failed (critical): {N}
- Failed (warning): {N}
- Skipped (irrelevant): {N}
### Critical Issues (must fix before deploy)
1. **{Check ID} — {Title}** [severity: critical]
- Current state: {what you found}
- Expected: {what should be}
- Fix: {exact action}
### Warnings (should fix)
1. **{Check ID} — {Title}** [severity: warning]
- Current state: {what you found}
- Expected: {what should be}
- Fix: {exact action}
### Skipped (irrelevant for this site type)
1. **{Check ID} — {Title}** [reason: {why skipped}]
### Verification Commands
```bash
# Run these to verify fixes
curl -s https://<site>/llms.txt | head -n 5
curl -s https://<site>/agent | jq '.identity.name'
curl -s https://<site>/ | grep -o 'application/ld+json' | wc -l
## Quick Verification Script
For local dev server (replace `http://localhost:3000`):
```bash
#!/bin/bash
BASE="http://localhost:3000"
echo "=== AEO Quick Check ==="
echo "--- Sitemap ---"
curl -sf "$BASE/sitemap.xml" > /dev/null && echo "PASS" || echo "FAIL: sitemap.xml"
echo "--- robots.txt ---"
curl -sf "$BASE/robots.txt" > /dev/null && echo "PASS" || echo "FAIL: robots.txt"
echo "--- llms.txt ---"
curl -sf "$BASE/llms.txt" | head -1 | grep "^# " > /dev/null && echo "PASS" || echo "FAIL: llms.txt"
echo "--- llms-full.txt ---"
curl -sf "$BASE/llms-full.txt" | wc -c | awk '{print ($1 > 2000 ? "PASS" : "FAIL: too short")}'
echo "--- index.md ---"
curl -sf "$BASE/index.md" | head -1 | grep "^# " > /dev/null && echo "PASS" || echo "FAIL: index.md"
echo "--- agent JSON ---"
curl -sf "$BASE/agent" | jq -e '.identity' > /dev/null 2>&1 && echo "PASS" || echo "FAIL: /agent"
echo "--- mode=agent ---"
curl -sf "$BASE/?mode=agent" | grep -q '<h1>' && echo "PASS" || echo "FAIL: ?mode=agent"
echo "--- JSON-LD ---"
curl -sf "$BASE/" | grep -q 'application/ld+json' && echo "PASS" || echo "FAIL: no JSON-LD"
echo "--- h1 in raw HTML ---"
curl -sf "$BASE/" | grep -q '<h1' && echo "PASS" || echo "FAIL: no h1"
echo "--- Link header ---"
curl -sI "$BASE/" | grep -qi 'Link:.*llms.txt' && echo "PASS" || echo "FAIL: no Link header"
<body> not <head>. Search engines and agents parse both, but <body> is more reliable for dynamic injection.content/site.ts) rather than hardcoding. If hardcoding is unavoidable, document it in agents.md.