#!/usr/bin/env bash
# AEO 上线后自查脚本 —— 对照 Osmani 六层 (addyosmani.com/blog/agentic-engine-optimization/)
# 用法: bash aeo-check.sh example.com [docs路径, 默认 /]
# 只读、只发 GET；输出每项 PASS/FAIL/WARN + 原始证据。不接受形容词，判据全在下面写死。
set -u
DOMAIN="${1:?用法: aeo-check.sh <domain> [docs-path]}"
DOCS="${2:-/}"
BASE="https://${DOMAIN}"
UA_LIST="GPTBot ChatGPT-User OAI-SearchBot ClaudeBot Claude-User Claude-SearchBot anthropic-ai Google-Extended PerplexityBot Perplexity-User"
TMP="$(mktemp -d)"
c() { curl -sL --max-time 20 "$@"; }
# 存在性判定：200 且 content-type 不是 text/html（SPA 兜底页对任意路径都回 200，是假阳性）
exists() { local ct; ct=$(c -o /dev/null -w '%{http_code} %{content_type}' "$1"); case "$ct" in 200\ *text/html*) echo "假阳性(200 但 HTML 兜底)";; 200*) echo "存在 [$ct]";; *) echo "无 [${ct%% *}]";; esac; }
tok() { python -c "import sys;print(len(sys.stdin.read())//4)"; }   # Osmani: chars/4 粗估

echo "== AEO check: ${BASE}  $(date -u +%FT%TZ) =="

# L1 robots.txt
echo; echo "[L1 robots.txt]"
RC=$(c -o "$TMP/robots" -w '%{http_code}' "$BASE/robots.txt")
if [ "$RC" = "200" ]; then
  echo "  robots.txt: 200 ($(wc -c <"$TMP/robots") bytes)"
  for ua in $UA_LIST; do
    if grep -qi "User-agent: *${ua}" "$TMP/robots" && awk -v ua="$ua" 'BEGIN{IGNORECASE=1} $0 ~ "User-agent: *"ua {f=1;next} /^User-agent:/{f=0} f && /Disallow: *\/ *$/ {found=1} END{exit !found}' "$TMP/robots"; then
      echo "  FAIL  ${ua} 被整站 Disallow"
    fi
  done
  awk 'BEGIN{IGNORECASE=1} /User-agent: *\*/{f=1;next} /^User-agent:/{f=0} f && /Disallow: *\/ *$/ {print "  FAIL  User-agent:* 整站 Disallow（所有 agent 都进不来）"}' "$TMP/robots"
  echo "  PASS  (无上列 AI UA 被整站封锁即通过；名单需对照各家官方 UA 文档更新)"
else
  echo "  WARN  robots.txt HTTP ${RC}（无文件=默认放行，但也没声明许可）"
fi
# agent-permissions.json（Osmani 原文有、极客公园版没有）
echo "  agent-permissions.json: $(exists "$BASE/agent-permissions.json") (可选)"

# L2 llms.txt
echo; echo "[L2 llms.txt]"
RC=$(c -o "$TMP/llms" -w '%{http_code}' "$BASE/llms.txt")
if [ "$RC" = "200" ]; then
  T=$(tok <"$TMP/llms"); N=$(grep -cE '^\s*-\s*\[' "$TMP/llms")
  echo "  llms.txt: 200, ~${T} tokens, ${N} 条目"
  [ "$T" -le 5000 ] && echo "  PASS  ≤5000 tokens" || echo "  FAIL  索引本身 >5000 tokens"
  grep -qiE '[0-9]+ ?k? ?tokens' "$TMP/llms" && echo "  PASS  条目带 token 数" || echo "  WARN  条目未标 token 数"
else
  echo "  FAIL  llms.txt HTTP ${RC}"
fi

# L3 skill.md
echo; echo "[L3 skill.md]"
FOUND=0
for p in /skill.md /.well-known/skill.md /docs/skill.md /SKILL.md; do
  RC=$(c -o "$TMP/skill" -w '%{http_code}' "$BASE$p")
  if [ "$RC" = "200" ] && grep -qiE '^(#|---|name:)' "$TMP/skill"; then
    FOUND=1; echo "  skill.md: 200 at $p (~$(tok <"$TMP/skill") tokens)"
    for sec in "can accomplish|能做|capabilit" "input|输入" "constraint|limit|限制" "doc|文档"; do
      grep -qiE "$sec" "$TMP/skill" && echo "  PASS  含 [$sec]" || echo "  FAIL  缺 [$sec]"
    done; break
  fi
done
[ $FOUND = 0 ] && echo "  FAIL  未找到 skill.md（试过 /skill.md /.well-known/skill.md /docs/skill.md）"

# L4 内容可解析：Markdown 直达 + 无 JS 也有正文
echo; echo "[L4 content]"
PAGE="$BASE$DOCS"
c -A "Mozilla/5.0" -o "$TMP/page.html" "$PAGE"
HB=$(wc -c <"$TMP/page.html")
TXT=$(python - "$TMP/page.html" <<'PY'
import re,sys,html
s=open(sys.argv[1],encoding='utf-8',errors='ignore').read()
s=re.sub(r'<script.*?</script>|<style.*?</style>|<nav.*?</nav>|<footer.*?</footer>','',s,flags=re.S|re.I)
s=html.unescape(re.sub(r'<[^>]+>',' ',s)); s=re.sub(r'\s+',' ',s).strip()
print(len(s))
PY
)
echo "  ${PAGE}: HTML ${HB} bytes, 去标签后正文 ${TXT} chars (~$((TXT/4)) tokens)"
[ "$TXT" -ge 400 ] && echo "  PASS  不执行 JS 也有正文（≥400 chars）" || echo "  FAIL  无 JS 正文≈空，疑 SPA 渲染 → agent 拿到空页"
[ "$HB" -gt 0 ] && echo "  HTML/正文 膨胀比: $(python -c "print(round($HB/max($TXT,1),1))")x  (>3x 说明标签噪音重，需提供 .md)"
for md in "${DOCS%/}.md" "${DOCS%/}/index.md" "${DOCS}?format=md"; do
  RC=$(c -o "$TMP/md" -w '%{http_code}' "$BASE$md")
  if [ "$RC" = "200" ] && head -c 2000 "$TMP/md" | grep -qE '^#|^\- |^\* '; then echo "  PASS  Markdown 直达: $md"; MDOK=1; break; fi
done
[ "${MDOK:-0}" = 0 ] && echo "  FAIL  无 .md 直达路径"
PT=$((TXT/4))
[ "$PT" -le 30000 ] && echo "  PASS  单页 ≤30K tokens (Osmani 硬上限)" || echo "  FAIL  单页 ${PT} tokens >30K，需分块"

# L5 token 元数据
echo; echo "[L5 token metadata]"
H=$(c -I -A "Mozilla/5.0" "$PAGE" | grep -iE '^x-token|token-count' || true)
M=$(grep -ioE '<meta[^>]*token[^>]*>' "$TMP/page.html" | head -3 || true)
{ [ -n "$H" ] || [ -n "$M" ]; } && { echo "  PASS"; echo "  $H $M"; } || echo "  FAIL  无 token 数 header/meta（实现: chars/4 → <meta name=token-count> 或 X-Token-Count）"

# L6 Copy for AI / AGENTS.md / MCP
echo; echo "[L6 UX bridge]"
grep -qiE 'copy (for|to) ai|copy as markdown' "$TMP/page.html" && echo "  PASS  Copy for AI 按钮" || echo "  WARN  未见 Copy for AI 按钮（低成本项）"
echo "  AGENTS.md: $(exists "$BASE/AGENTS.md") (仓库有则应有)"
echo "  /mcp: $(exists "$BASE/mcp") (可选，Osmani 清单项；MCP 端点通常回 JSON/SSE 或 4xx，HTML=兜底)"

# Agent 视角首屏：前 500 tokens 能否答「是什么/能干什么/怎么开始」—— 机器只能给你看，判断你自己做
echo; echo "[首 500 tokens 原文 —— 人工判三问]"
python - "$TMP/page.html" <<'PY'
import re,sys,html
s=open(sys.argv[1],encoding='utf-8',errors='ignore').read()
s=re.sub(r'<script.*?</script>|<style.*?</style>|<nav.*?</nav>|<footer.*?</footer>','',s,flags=re.S|re.I)
s=html.unescape(re.sub(r'<[^>]+>',' ',s)); s=re.sub(r'\s+',' ',s).strip()
print("  "+s[:2000])
PY
rm -rf "$TMP"