kiku-agent 1.3.2 → 1.3.4

package/server/agent/webSearch.js

@@ -3,17 +3,20 @@ import { execFile } from 'node:child_process';
 import { mkdtemp, readFile, rm } from 'node:fs/promises';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
+import { resolveDynamicWindowsAppPaths } from './appPaths.js';
 import http from 'node:http';
 import https from 'node:https';
 import { lookup as dnsLookup } from 'node:dns/promises';
 import { assertSafeUrlHostname, sanitizeUrl, isPrivateIPv4, isPrivateIPv6, isIpAddress } from './sanitizer.js';
-import { getContextualizedSearchQuery } from './catalogEntityResolver.js';
+import { getContextualizedSearchQuery, isQueryInDomain, THAI_INTENT_MAP } from './catalogEntityResolver.js';
+import { dedupeNewsItems, rankNewsItems } from './newsIntelligence.js';
 import { galleryData } from '../../src/data/galleryData.js';
 import { discographyData } from '../../src/data/discographyData.js';
 import { musicVideoData } from '../../src/data/musicVideoData.js';
 import { profileData } from '../../src/data/profileData.js';
 import { defaultMetaSearch, hybridWebSearch } from './websearch/index.js';
 import { computeSearchCacheKey, TTL_GENERAL_MS } from './websearch/resilience/cache.js';
+import { appendDomainFilters, planSearchRequest, rerankSearchResults } from './searchQueryPlanner.js';
 import {
   decodeHtmlBytes,
   decodeEntities,
@@ -156,14 +159,17 @@ export async function safeFetch(inputUrl, options = {}, maxRedirects = 5) {
   let currentOptions = { ...options };
 
   for (let redirects = 0; redirects <= maxRedirects; redirects++) {
-    const sanitized = sanitizeUrl(currentUrl);
+    const sanitized = sanitizeUrl(currentUrl, { allowLoopback: Boolean(currentOptions.allowLoopback) });
     if (!sanitized.safe) throw new Error(`URL rejected by SSRF guard: ${sanitized.reason}`);
 
     const parsedUrl = sanitized.url;
     const hostname = sanitized.hostname;
+    const isLoopback = hostname === 'localhost' || hostname === '127.0.0.1' || hostname === '::1';
 
     let targetIp = hostname;
-    if (!isIpAddress(hostname)) {
+    if (currentOptions.allowLoopback && isLoopback) {
+      targetIp = hostname;
+    } else if (!isIpAddress(hostname)) {
       const addresses = await safeFetchDnsLookup(hostname);
       if (!addresses || addresses.length === 0) {
         throw new Error(`Unable to resolve hostname ${hostname}`);
@@ -603,14 +609,47 @@ async function searchBing(query, maxResults, { language } = {}) {
   return results;
 }
 
+function prepareSearchQuery(rawQuery) {
+  let q = String(rawQuery || '').trim();
+  if (!q) return '';
+  for (const [regex, replacement] of THAI_INTENT_MAP) {
+    if (regex.test(q)) {
+      q = `${q} ${replacement}`;
+      break;
+    }
+  }
+  return q;
+}
+
+function buildQueryVariants(rawQuery) {
+  const q = String(rawQuery || '').trim();
+  const variants = [];
+  const noQuotes = q.replace(/["'«»“”„]/g, ' ').replace(/\s+/g, ' ').trim();
+  if (noQuotes && noQuotes !== q) variants.push(noQuotes);
+
+  const cleanStopwords = noQuotes.replace(/\b(what|where|who|is|are|tell me|about|details|news|ขอข้อมูล|เกี่ยวกับ|คือใคร|มีอะไรบ้าง|ไหม|มั้ย)\b/gi, ' ').replace(/\s+/g, ' ').trim();
+  if (cleanStopwords && cleanStopwords !== noQuotes && cleanStopwords.length > 2) {
+    variants.push(cleanStopwords);
+  }
+  return variants;
+}
+
 export async function searchWeb(query, maxResults = 8, options = {}) {
   if (!query || typeof query !== 'string') {
     return { status: 'error', message: 'Query string is required for web search.' };
   }
 
-  const lang = detectLanguage(query);
+  const ownedPlan = !options.queryPlan;
+  const queryPlan = options.queryPlan || planSearchRequest({
+    keywords: query,
+    target_language: options.targetLanguage,
+    domain_filter: options.allowedDomains,
+  }, { kind: 'web' });
+  const originalQuery = options.originalQuery || queryPlan.originalQuery || query;
+  const fetchBase = ownedPlan ? (queryPlan.keywords || query) : query;
+  const lang = queryPlan.targetLanguage || detectLanguage(fetchBase);
   const capped = Math.min(Math.max(maxResults, 1), 25);
-  const cacheKey = computeSearchCacheKey('searchWeb', query, capped);
+  const cacheKey = computeSearchCacheKey('searchWeb', `${query}|${queryPlan.domains.join(',')}`, capped);
 
   if (!options.noCache) {
     const cached = defaultMetaSearch.cache.get(cacheKey);
@@ -619,68 +658,91 @@ export async function searchWeb(query, maxResults = 8, options = {}) {
     }
   }
 
-  try {
-    const [metaResSettled, imgResSettled] = await Promise.allSettled([
-      defaultMetaSearch.executeSearch(query, {
-        limit: capped,
-        language: lang,
-        ...options
-      }),
-      searchWebImages(query, 6).catch(() => null)
-    ]);
+  const preparedQuery = appendDomainFilters(options.skipPrep ? fetchBase : prepareSearchQuery(fetchBase), queryPlan.domains);
 
-    const metaRes = metaResSettled.status === 'fulfilled' ? metaResSettled.value : null;
-    const imgRes = imgResSettled.status === 'fulfilled' ? imgResSettled.value : null;
-    const images = Array.isArray(imgRes?.results) ? imgRes.results.slice(0, 6) : [];
+  const runAttempt = async (targetQuery) => {
+    try {
+      const [metaResSettled, imgResSettled] = await Promise.allSettled([
+        defaultMetaSearch.executeSearch(targetQuery, {
+          limit: capped,
+          language: lang,
+          ...options
+        }),
+        searchWebImages(targetQuery, 6, { queryPlan }).catch(() => null)
+      ]);
 
-    if (metaRes && metaRes.results && metaRes.results.length > 0) {
-      const sources = [...new Set(metaRes.results.map(r => r.engine || metaRes.engine || 'web'))];
-      const resultObj = {
-        status: 'success',
-        query,
-        language: lang,
-        count: metaRes.results.length,
-        sources,
-        results: metaRes.results.map(r => ({
+      const metaRes = metaResSettled.status === 'fulfilled' ? metaResSettled.value : null;
+      const imgRes = imgResSettled.status === 'fulfilled' ? imgResSettled.value : null;
+      const images = Array.isArray(imgRes?.results) ? imgRes.results.slice(0, 6) : [];
+
+      if (metaRes && metaRes.results && metaRes.results.length > 0) {
+        const rankedResults = rerankSearchResults(metaRes.results.map(r => ({
           title: r.title,
           url: r.link,
           snippet: r.snippet,
           source: r.engine || 'web'
-        })),
-        ...(images.length > 0 ? { images } : {}),
-        ...(metaRes.mode === 'deep_scrape' ? {
-          scrapedPages: metaRes.scrapedPages,
-          combinedContent: metaRes.combinedContent
-        } : {})
-      };
-      if (!options.noCache) {
-        defaultMetaSearch.cache.set(cacheKey, resultObj, TTL_GENERAL_MS);
+        })), queryPlan, { limit: capped });
+        const sources = [...new Set(rankedResults.map(r => r.source || 'web'))];
+        const resultObj = {
+          status: 'success',
+          query: originalQuery,
+          normalizedQuery: queryPlan.keywords,
+          language: lang,
+          count: rankedResults.length,
+          sources,
+          results: rankedResults,
+          ...(images.length > 0 ? { images } : {}),
+          ...(metaRes.mode === 'deep_scrape' ? {
+            scrapedPages: metaRes.scrapedPages,
+            combinedContent: metaRes.combinedContent
+          } : {})
+        };
+        if (!options.noCache) {
+          defaultMetaSearch.cache.set(cacheKey, resultObj, TTL_GENERAL_MS);
+        }
+        return resultObj;
       }
-      return resultObj;
-    }
-  } catch {}
+    } catch {}
 
-  try {
-    const wikiItems = await searchWikipedia(query, Math.min(capped, 3), lang);
-    if (wikiItems.length > 0) {
-      const wikiResponse = {
-        status: 'success',
-        query,
-        language: lang,
-        count: wikiItems.length,
-        sources: ['Wikipedia'],
-        results: wikiItems
-      };
-      if (!options.noCache) {
-        defaultMetaSearch.cache.set(cacheKey, wikiResponse, TTL_GENERAL_MS);
+    try {
+      const wikiItems = await searchWikipedia(targetQuery, Math.min(capped, 3), lang);
+      if (wikiItems.length > 0) {
+        const wikiResponse = {
+          status: 'success',
+          query: originalQuery,
+          normalizedQuery: queryPlan.keywords,
+          language: lang,
+          count: wikiItems.length,
+          sources: ['Wikipedia'],
+          results: wikiItems
+        };
+        if (!options.noCache) {
+          defaultMetaSearch.cache.set(cacheKey, wikiResponse, TTL_GENERAL_MS);
+        }
+        return wikiResponse;
       }
-      return wikiResponse;
-    }
-  } catch {}
+    } catch {}
+    return null;
+  };
+
+  const primaryResult = await runAttempt(preparedQuery);
+  if (primaryResult) return primaryResult;
+
+  if (preparedQuery !== query) {
+    const rawResult = await runAttempt(query);
+    if (rawResult) return rawResult;
+  }
+
+  const variants = buildQueryVariants(query);
+  for (const variant of variants) {
+    const variantResult = await runAttempt(variant);
+    if (variantResult) return variantResult;
+  }
 
   return {
     status: 'no_results',
-    query,
+    query: originalQuery,
+    normalizedQuery: queryPlan.keywords,
     language: lang,
     count: 0,
     sources: [],
@@ -691,11 +753,31 @@ export async function searchWeb(query, maxResults = 8, options = {}) {
 const PAGE_CACHE_TTL_MS = 10 * 60 * 1000;
 const pageContentCache = new BoundedLruCache(40);
 
-function buildPageResult(pageData, { finalUrl, maxLength, query, cached }) {
+export function buildPageResult(pageData, { finalUrl, maxLength, query, cached }) {
   const markdown = pageData.markdown || '';
+  const totalChars = markdown.length;
+
+  if (!markdown.trim()) {
+    return {
+      status: 'error',
+      url: finalUrl,
+      contentType: 'html',
+      contentLength: 0,
+      totalPageChars: 0,
+      message: 'No readable text could be extracted from this page (login-walled or JavaScript-rendered site such as Weibo, Douyin or Xiaohongshu). Do NOT present this as the final answer — retry in the same turn through an alternate route: (1) the mobile mirror of the URL (e.g. weibo.com → m.weibo.cn keeping the same path), (2) search_chinese_web with the topic keywords, (3) search_news on the topic, or (4) web_search for corroborating pages. Only tell the user the page is unavailable after an alternate route also returns nothing.',
+      meta: pageData.meta || {},
+      extraction: {
+        engine: pageData.engine || 'cheerio-readability',
+        mode: 'no_content',
+        cached: Boolean(cached),
+        charset: pageData.charset || 'utf-8',
+        quality: pageData.quality || null
+      }
+    };
+  }
+
   let content = markdown;
   let contentMode = 'full';
-  const totalChars = markdown.length;
 
   if (query) {
     const passages = splitIntoPassages(markdown);
@@ -718,7 +800,7 @@ function buildPageResult(pageData, { finalUrl, maxLength, query, cached }) {
     contentType: 'html',
     contentLength: content.length,
     totalPageChars: totalChars,
-    content: content || 'Unable to retrieve text content from this webpage.',
+    content,
     meta: pageData.meta || {},
     extraction: {
       engine: pageData.engine || 'cheerio-readability',
@@ -964,7 +1046,7 @@ export async function fetchJson(url, options = {}) {
   } catch {
     return { status: 'error', message: 'Invalid URL format.' };
   }
-  const ssrf = await assertSafeUrlHostname(parsedUrl.hostname);
+  const ssrf = await assertSafeUrlHostname(parsedUrl.hostname, { allowLoopback: Boolean(options.allowLoopback) });
   if (!ssrf.safe) {
     return { status: 'error', message: `URL rejected by SSRF guard: ${ssrf.reason}` };
   }
@@ -988,7 +1070,8 @@ export async function fetchJson(url, options = {}) {
         'User-Agent': 'KikuAgent/2.0',
         ...headers
       },
-      signal: controller.signal
+      signal: controller.signal,
+      allowLoopback: Boolean(options.allowLoopback)
     };
 
     if (body && ['POST', 'PUT', 'PATCH'].includes(fetchOptions.method)) {
@@ -1056,13 +1139,7 @@ export async function fetchJson(url, options = {}) {
 function findChromePath() {
   const p = process.platform;
   if (p === 'win32') {
-    const candidates = [
-      (process.env.LOCALAPPDATA || '') + '/Google/Chrome/Application/chrome.exe',
-      (process.env['PROGRAMFILES'] || '') + '/Google/Chrome/Application/chrome.exe',
-      (process.env['PROGRAMFILES(X86)'] || '') + '/Google/Chrome/Application/chrome.exe',
-      'C:/Program Files/Google/Chrome/Application/chrome.exe',
-      'C:/Program Files (x86)/Google/Chrome/Application/chrome.exe'
-    ];
+    const candidates = resolveDynamicWindowsAppPaths('Google', 'Chrome', 'Application', 'chrome.exe');
     for (const c of candidates) {
       if (c && existsSync(c)) return c;
     }
@@ -1303,69 +1380,88 @@ async function fetchYahooNewsRss(query, maxResults) {
   }
 }
 
-export async function searchNews(query, maxResults = 8) {
+export async function searchNews(query, maxResults = 8, options = {}) {
   if (!query || typeof query !== 'string') {
     return { status: 'error', message: 'Query string is required for news search.' };
   }
 
+  const ownedPlan = !options.queryPlan;
+  const queryPlan = options.queryPlan || planSearchRequest(query, { kind: 'news' });
+  const searchBase = ownedPlan ? (queryPlan.keywords || query) : query;
   const capped = Math.min(Math.max(maxResults, 1), 20);
-  const lang = detectLanguage(query);
-  const results = [];
+  const lang = queryPlan.targetLanguage || detectLanguage(searchBase);
   const sources = [];
-  const attempts = [];
   const seenUrls = new Set();
+  const rawItems = [];
 
   const add = (item, defaultSource) => {
     if (!item?.url || seenUrls.has(item.url)) return;
     seenUrls.add(item.url);
     const src = item.source || defaultSource;
-    results.push({ ...item, source: src });
+    rawItems.push({ ...item, source: src });
     if (!sources.includes(src)) sources.push(src);
   };
 
-  const [gSettled, bSettled, ySettled, ddgSettled, imgSettled] = await Promise.allSettled([
-    fetchGoogleNewsRss(query, capped, lang),
-    fetchBingNewsRss(query, capped, lang),
-    fetchYahooNewsRss(query, Math.ceil(capped / 2)),
-    searchDuckDuckGo(`${query} news`, Math.min(capped, 4)),
-    searchWebImages(`${query} news`, 6).catch(() => null)
-  ]);
-
-  if (gSettled.status === 'fulfilled' && Array.isArray(gSettled.value)) {
-    attempts.push({ source: 'Google News', status: gSettled.value.length ? 'success' : 'no_results', count: gSettled.value.length });
-  } else {
-    attempts.push({ source: 'Google News', status: 'error', message: gSettled.reason?.message });
+  const isArtistQuery = /(?:จวีจิ้งอี|จิ้งอี|ju\s*jing\s*yi|kiku|鞠婧祎|小鞠)/i.test(`${query} ${searchBase}`);
+  const cleanLocalQuery = searchBase
+    .replace(/(?:ข่าวล่าสุด|ข่าวใหม่|ข่าวสาร|ข่าว|ล่าสุด|เมื่อเร็วๆ\s*นี้|update|news|when:\d+d)/gi, ' ')
+    .replace(/\s+/g, ' ')
+    .trim() || searchBase;
+  const filteredLocalQuery = appendDomainFilters(cleanLocalQuery, queryPlan.domains);
+
+  const cnQuery = isArtistQuery ? getContextualizedSearchQuery(searchBase, 'chinese_web') : null;
+
+  const tasks = [
+    fetchGoogleNewsRss(filteredLocalQuery, capped, lang),
+    fetchBingNewsRss(filteredLocalQuery, capped, lang),
+    fetchYahooNewsRss(filteredLocalQuery, Math.ceil(capped / 2)),
+    searchDuckDuckGo(filteredLocalQuery, Math.min(capped, 4)),
+    searchWebImages(`${filteredLocalQuery} news`, 6, { queryPlan }).catch(() => null)
+  ];
+
+  if (cnQuery && lang !== 'zh') {
+    tasks.push(
+      fetchGoogleNewsRss(cnQuery, capped, 'zh').catch(() => []),
+      fetchBingNewsRss(cnQuery, capped, 'zh').catch(() => [])
+    );
   }
 
-  if (bSettled.status === 'fulfilled' && Array.isArray(bSettled.value)) {
-    attempts.push({ source: 'Bing News', status: bSettled.value.length ? 'success' : 'no_results', count: bSettled.value.length });
-  } else {
-    attempts.push({ source: 'Bing News', status: 'error', message: bSettled.reason?.message });
-  }
+  const settled = await Promise.allSettled(tasks);
+
+  const gList = (settled[0]?.status === 'fulfilled' && Array.isArray(settled[0].value)) ? settled[0].value : [];
+  const bList = (settled[1]?.status === 'fulfilled' && Array.isArray(settled[1].value)) ? settled[1].value : [];
+  const yList = (settled[2]?.status === 'fulfilled' && Array.isArray(settled[2].value)) ? settled[2].value : [];
+  const ddgList = (settled[3]?.status === 'fulfilled' && Array.isArray(settled[3].value)) ? settled[3].value : [];
+  const imgRes = settled[4]?.status === 'fulfilled' ? settled[4].value : null;
+  const gCnList = (settled[5]?.status === 'fulfilled' && Array.isArray(settled[5].value)) ? settled[5].value : [];
+  const bCnList = (settled[6]?.status === 'fulfilled' && Array.isArray(settled[6].value)) ? settled[6].value : [];
 
-  const gList = (gSettled.status === 'fulfilled' && Array.isArray(gSettled.value)) ? gSettled.value : [];
-  const bList = (bSettled.status === 'fulfilled' && Array.isArray(bSettled.value)) ? bSettled.value : [];
-  const yList = (ySettled.status === 'fulfilled' && Array.isArray(ySettled.value)) ? ySettled.value : [];
-  const ddgList = (ddgSettled.status === 'fulfilled' && Array.isArray(ddgSettled.value)) ? ddgSettled.value : [];
-  const imgRes = imgSettled.status === 'fulfilled' ? imgSettled.value : null;
   const newsImages = Array.isArray(imgRes?.results) ? imgRes.results.slice(0, 6) : [];
 
   let i = 0;
-  while (results.length < capped && (i < gList.length || i < bList.length || i < yList.length || i < ddgList.length)) {
-    if (i < gList.length && results.length < capped) add(gList[i], 'Google News');
-    if (i < bList.length && results.length < capped) add(bList[i], 'Bing News');
-    if (i < yList.length && results.length < capped) add(yList[i], 'Yahoo News');
-    if (i < ddgList.length && results.length < capped) add(ddgList[i], 'Web News');
+  const maxLen = Math.max(gList.length, bList.length, yList.length, ddgList.length, gCnList.length, bCnList.length);
+  while (i < maxLen) {
+    if (i < gList.length) add(gList[i], 'Google News');
+    if (i < bList.length) add(bList[i], 'Bing News');
+    if (i < gCnList.length) add(gCnList[i], 'Google News (CN)');
+    if (i < bCnList.length) add(bCnList[i], 'Bing News (CN)');
+    if (i < yList.length) add(yList[i], 'Yahoo News');
+    if (i < ddgList.length) add({ ...ddgList[i], url: ddgList[i].url || ddgList[i].link }, 'Web News');
     i++;
   }
 
+  const deduped = dedupeNewsItems(rawItems);
+  const ranked = rankNewsItems(deduped);
+  const finalResults = rerankSearchResults(ranked, queryPlan, { limit: capped });
+
   return {
-    status: results.length > 0 ? 'success' : 'no_results',
-    query,
+    status: finalResults.length > 0 ? 'success' : 'no_results',
+    query: queryPlan.originalQuery || query,
+    normalizedQuery: queryPlan.keywords,
     language: lang,
-    count: results.length,
+    count: finalResults.length,
     sources,
-    results,
+    results: finalResults,
     ...(newsImages.length > 0 ? { images: newsImages } : {})
   };
 }
@@ -1869,6 +1965,23 @@ export function parseBaiduResults(html, maxResults = 8) {
       results.push({ title, url: match[1], snippet, source: 'Baidu' });
     }
   }
+
+  if (results.length === 0) {
+    const altRegex = /<(?:a[^>]+href="([^"]+)"[^>]*>[\s\S]*?<h3[^>]*>([\s\S]*?)<\/h3>|<h3[^>]*>[\s\S]*?<a[^>]+href="([^"]+)"[^>]*>([\s\S]*?)<\/a>|<div[^>]+class="[^"]*c-title[^"]*"[^>]*>[\s\S]*?<a[^>]+href="([^"]+)"[^>]*>([\s\S]*?)<\/a>)([\s\S]*?)(?=<article|<div[^>]+class="[^"]*c-result|<h3|<\/body>|$)/gi;
+    while ((match = altRegex.exec(html)) !== null && results.length < maxResults) {
+      const url = match[1] || match[3] || match[5];
+      const rawTitle = match[2] || match[4] || match[6];
+      const afterText = match[7] || '';
+      if (!url || !rawTitle) continue;
+      const title = cleanHtml(rawTitle);
+      const snippet = cleanHtml(afterText).replace(/["']?\}[\],\}\s]+.*$/g, '').trim().slice(0, 320);
+      if (title && url && !seenUrls.has(url)) {
+        seenUrls.add(url);
+        results.push({ title, url, snippet, source: 'Baidu' });
+      }
+    }
+  }
+
   return results;
 }
 
@@ -1877,9 +1990,13 @@ export async function searchChineseWeb(query, maxResults = 8, options = {}) {
     return { status: 'error', message: 'Query string is required for Chinese web search.' };
   }
 
-  const effectiveQuery = typeof options.queryEnricher === 'function'
-    ? options.queryEnricher(query, 'chinese_web')
-    : getContextualizedSearchQuery(query, 'chinese_web');
+  const ownedPlan = !options.queryPlan;
+  const queryPlan = options.queryPlan || planSearchRequest({ keywords: query, domain_filter: options.allowedDomains }, { kind: 'chinese' });
+  const fetchBase = ownedPlan ? (queryPlan.keywords || query) : query;
+  const enrichedQuery = typeof options.queryEnricher === 'function'
+    ? options.queryEnricher(fetchBase, 'chinese_web')
+    : getContextualizedSearchQuery(fetchBase, 'chinese_web');
+  const effectiveQuery = appendDomainFilters(enrichedQuery, queryPlan.domains);
 
   const capped = Math.min(Math.max(maxResults, 1), 20);
   const results = [];
@@ -1896,21 +2013,51 @@ export async function searchChineseWeb(query, maxResults = 8, options = {}) {
           'Accept-Language': 'zh-CN,zh;q=0.9',
           'Referer': 'https://www.baidu.com/'
         }
-      }, 10000);
+      }, 4500);
       if (res.ok) {
         const html = await readResponseTextLimited(res, 3_000_000);
-        if (html.includes('百度安全验证') || html.includes('wappass.baidu.com')) {
+        if (!html.includes('百度安全验证') && !html.includes('wappass.baidu.com')) {
+          const items = parseBaiduResults(html, capped);
+          if (items.length > 0) {
+            attempts.push({ source: 'Baidu', status: 'success', count: items.length, durationMs: Date.now() - t0 });
+            return items;
+          }
+        } else {
           attempts.push({ source: 'Baidu', status: 'blocked', message: 'Baidu anti-bot verification challenge.', durationMs: Date.now() - t0 });
-          return [];
         }
-        const items = parseBaiduResults(html, capped);
-        attempts.push({ source: 'Baidu', status: items.length ? 'success' : 'no_results', count: items.length, durationMs: Date.now() - t0 });
-        return items;
+      } else {
+        attempts.push({ source: 'Baidu', status: 'error', message: `HTTP ${res.status}`, durationMs: Date.now() - t0 });
       }
-      attempts.push({ source: 'Baidu', status: 'error', message: `HTTP ${res.status}`, durationMs: Date.now() - t0 });
     } catch (err) {
-      attempts.push({ source: 'Baidu', status: 'error', message: err.name === 'AbortError' ? 'Baidu timeout (10s)' : err.message, durationMs: Date.now() - t0 });
+      attempts.push({ source: 'Baidu', status: 'error', message: err.name === 'AbortError' ? 'Baidu timeout (4.5s)' : err.message, durationMs: Date.now() - t0 });
+    }
+
+    // Mobile Baidu fallback (less prone to desktop captcha challenges)
+    try {
+      const tMobile = Date.now();
+      const mobileUrl = `https://m.baidu.com/s?word=${encodeURIComponent(effectiveQuery)}`;
+      const resM = await fetchWithTimeout(mobileUrl, {
+        headers: {
+          ...BROWSER_HEADERS,
+          'User-Agent': 'Mozilla/5.0 (iPhone; CPU iPhone OS 17_0 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Mobile/15E148 Safari/604.1',
+          'Accept-Language': 'zh-CN,zh;q=0.9',
+          'Referer': 'https://m.baidu.com/'
+        }
+      }, 3500);
+      if (resM.ok) {
+        const htmlM = await readResponseTextLimited(resM, 3_000_000);
+        if (!htmlM.includes('百度安全验证') && !htmlM.includes('wappass.baidu.com')) {
+          const mItems = parseBaiduResults(htmlM, capped);
+          if (mItems.length > 0) {
+            attempts.push({ source: 'Baidu (Mobile)', status: 'success', count: mItems.length, durationMs: Date.now() - tMobile });
+            return mItems;
+          }
+        }
+      }
+    } catch (_mErr) {
+      // Graceful mobile fallback failure
     }
+
     return [];
   };
 
@@ -2012,7 +2159,7 @@ export async function searchChineseWeb(query, maxResults = 8, options = {}) {
   const fetchOpenWeb = async () => {
     const t0 = Date.now();
     try {
-      const cnDomains = options.site ? [options.site] : ['sina.com.cn', 'sohu.com', 'weibo.com', '163.com', 'qq.com'];
+      const cnDomains = options.site ? [options.site] : (queryPlan.domains.length > 0 ? queryPlan.domains : ['sina.com.cn', 'sohu.com', 'weibo.com', '163.com', 'qq.com']);
       const ddgPromises = cnDomains.slice(0, 3).map(domain =>
         searchDuckDuckGo(`${effectiveQuery} site:${domain}`, 2)
           .then(items => items.map(it => ({ ...it, source: `Web (${domain})` })))
@@ -2078,13 +2225,15 @@ export async function searchChineseWeb(query, maxResults = 8, options = {}) {
     }
   } catch {}
 
+  const rankedResults = rerankSearchResults(results, queryPlan, { limit: capped });
   return {
     status: results.length > 0 ? 'success' : 'no_results',
-    query,
-    count: results.length,
+    query: queryPlan.originalQuery || query,
+    normalizedQuery: queryPlan.keywords,
+    count: rankedResults.length,
     sources: [...new Set(sources)],
     attempts,
-    results,
+    results: rankedResults,
     ...(cnImages.length > 0 ? { images: cnImages } : {})
   };
 }
@@ -2119,20 +2268,76 @@ export function decodeBaiduUrl(url) {
   return res;
 }
 
-export async function searchWebImages(query, maxResults = 8) {
+const ARTIST_IMAGE_TOKEN_RE = /(鞠婧祎|ju\s?jing\s?yi|jujingyi|จวีจิ้งอี|จิ้งอี|小鞠|kikuu?|蜜橘)/i;
+
+export function titleMentionsArtist(title) {
+  return ARTIST_IMAGE_TOKEN_RE.test(String(title || '').replace(/<[^>]*>/g, ''));
+}
+
+export function isLowValueImageUrl(url) {
+  const raw = String(url || '');
+  if (!raw) return true;
+  if (/^data:/i.test(raw)) return true;
+  if (/\.svg(?:[?#]|$)/i.test(raw)) return true;
+  return /\/(?:logo|icons?|favicon|avatar|emoji|sprite)[^/]*$/i.test(raw);
+}
+
+export function imageHostKey(url) {
+  try {
+    const parsed = new URL(url);
+    return parsed.host + parsed.pathname.toLowerCase();
+  } catch {
+    return String(url || '');
+  }
+}
+
+export function buildImageSearchQuery(query) {
+  if (!isQueryInDomain(query)) {
+    return String(query || '').trim();
+  }
+  const effective = getContextualizedSearchQuery(query, 'chinese_web');
+  if (ARTIST_IMAGE_TOKEN_RE.test(effective)) return effective;
+  return ('鞠婧祎 ' + effective).trim();
+}
+
+export async function searchWebImages(query, maxResults = 8, options = {}) {
   if (!query || typeof query !== 'string') {
     return { status: 'error', message: 'Query string is required for image search.' };
   }
 
-  const effectiveQuery = getContextualizedSearchQuery(query, 'chinese_web');
+  const ownedPlan = !options.queryPlan;
+  const queryPlan = options.queryPlan || planSearchRequest(query, { kind: 'image' });
+  const fetchBase = ownedPlan ? (queryPlan.keywords || query) : query;
+  const inDomain = isQueryInDomain(fetchBase);
+  const effectiveQuery = appendDomainFilters(buildImageSearchQuery(fetchBase), queryPlan.domains);
+  const requireArtistMatch = inDomain && ARTIST_IMAGE_TOKEN_RE.test(effectiveQuery);
   const capped = Math.min(Math.max(maxResults, 1), 25);
   const results = [];
   const sources = [];
   const seenUrls = new Set();
+  const seenImageKeys = new Set();
+
+  const matchesDomainFilter = (...urls) => {
+    if (queryPlan.domains.length === 0) return true;
+    return urls.some(candidate => {
+      try {
+        const host = new URL(candidate).hostname.replace(/^www\./, '');
+        return queryPlan.domains.some(domain => host === domain || host.endsWith(`.${domain}`));
+      } catch {
+        return false;
+      }
+    });
+  };
 
-  const addImage = (item, defaultSource) => {
+  const addImage = (item, defaultSource, { trusted = false } = {}) => {
     if (!item?.imageUrl || seenUrls.has(item.imageUrl)) return false;
+    if (isLowValueImageUrl(item.imageUrl)) return false;
+    const hostKey = imageHostKey(item.imageUrl);
+    if (seenImageKeys.has(hostKey)) return false;
+    if (item.width && item.width < 200) return false;
+    if (requireArtistMatch && !trusted && !titleMentionsArtist(item.title)) return false;
     seenUrls.add(item.imageUrl);
+    seenImageKeys.add(hostKey);
     const src = item.source || defaultSource;
     let sourcePageUrl = item.sourcePageUrl || '';
     if (sourcePageUrl && (sourcePageUrl.includes('ippr_z2C$q') || sourcePageUrl.includes('AzdH3F'))) {
@@ -2141,19 +2346,23 @@ export async function searchWebImages(query, maxResults = 8) {
     if (sourcePageUrl && !/^https?:\/\//i.test(sourcePageUrl)) {
       sourcePageUrl = '';
     }
+    if (!matchesDomainFilter(sourcePageUrl, item.imageUrl)) return false;
     results.push({ ...item, sourcePageUrl, source: src });
     if (!sources.includes(src)) sources.push(src);
     return true;
   };
 
  
-  if (Array.isArray(galleryData)) {
-    const qLower = query.toLowerCase();
-    const localMatches = galleryData.filter(g =>
-      (g.title && g.title.toLowerCase().includes(qLower)) ||
-      (g.category && g.category.toLowerCase().includes(qLower)) ||
-      (g.tags && g.tags.some(t => t.toLowerCase().includes(qLower)))
-    );
+  if (inDomain && Array.isArray(galleryData)) {
+    const matchBases = [...new Set([fetchBase, queryPlan.originalQuery].map(v => String(v || '').toLowerCase()).filter(Boolean))];
+    const localMatches = galleryData.filter(g => {
+      const haystacks = [
+        (g.title && g.title.toLowerCase()) || '',
+        (g.category && g.category.toLowerCase()) || '',
+        ...((g.tags || []).map(t => String(t).toLowerCase())),
+      ];
+      return matchBases.some(base => haystacks.some(hay => hay && (hay.includes(base) || base.split(/\s+/).some(token => token.length > 1 && hay.includes(token)))));
+    });
     for (const g of localMatches.slice(0, 3)) {
       addImage({
         title: g.title || effectiveQuery,
@@ -2163,14 +2372,14 @@ export async function searchWebImages(query, maxResults = 8) {
         width: 1080,
         height: 1440,
         source: 'Official Gallery'
-      }, 'Official Gallery');
+      }, 'Official Gallery', { trusted: true });
     }
   }
 
 
-  const fetchBaiduImages = async () => {
+  const fetchBaiduImages = async (searchQuery = effectiveQuery) => {
     try {
-      const baiduImgUrl = `https://image.baidu.com/search/acjson?tn=resultjson_com&logid=${Date.now()}&ipn=rj&ct=201326592&is=&fp=result&word=${encodeURIComponent(effectiveQuery)}&queryWord=${encodeURIComponent(effectiveQuery)}&cl=2&lm=-1&ie=utf-8&oe=utf-8&pn=0&rn=${Math.min(capped * 2, 30)}&gsm=1e`;
+      const baiduImgUrl = `https://image.baidu.com/search/acjson?tn=resultjson_com&logid=${Date.now()}&ipn=rj&ct=201326592&is=&fp=result&word=${encodeURIComponent(searchQuery)}&queryWord=${encodeURIComponent(searchQuery)}&cl=2&lm=-1&ie=utf-8&oe=utf-8&pn=0&rn=${Math.min(capped * 2, 30)}&gsm=1e`;
       const res = await fetchWithTimeout(baiduImgUrl, {
         headers: {
           ...BROWSER_HEADERS,
@@ -2224,9 +2433,9 @@ export async function searchWebImages(query, maxResults = 8) {
   };
 
 
-  const fetchBingImages = async () => {
+  const fetchBingImages = async (searchQuery = effectiveQuery) => {
     try {
-      const url = `https://www.bing.com/images/async?q=${encodeURIComponent(effectiveQuery)}&mmasync=1`;
+      const url = `https://www.bing.com/images/async?q=${encodeURIComponent(searchQuery)}&mmasync=1`;
       const res = await fetchWithTimeout(url, {
         headers: {
           ...BROWSER_HEADERS,
@@ -2261,9 +2470,9 @@ export async function searchWebImages(query, maxResults = 8) {
   };
 
 
-  const fetchBilibiliCovers = async () => {
+  const fetchBilibiliCovers = async (searchQuery = effectiveQuery) => {
     try {
-      const url = `https://api.bilibili.com/x/web-interface/search/all/v2?keyword=${encodeURIComponent(effectiveQuery)}`;
+      const url = `https://api.bilibili.com/x/web-interface/search/all/v2?keyword=${encodeURIComponent(searchQuery)}`;
       const res = await fetchWithTimeout(url, {
         headers: {
           ...BROWSER_HEADERS,
@@ -2294,6 +2503,16 @@ export async function searchWebImages(query, maxResults = 8) {
   };
 
 
+  const mergeImageRound = (bList, biList, blList) => {
+    let idx = 0;
+    while (results.length < capped && (idx < bList.length || idx < biList.length || idx < blList.length)) {
+      if (idx < bList.length && results.length < capped) addImage(bList[idx], 'Baidu Images');
+      if (idx < biList.length && results.length < capped) addImage(biList[idx], 'Bing Images');
+      if (idx < blList.length && results.length < capped) addImage(blList[idx], 'Bilibili Covers');
+      idx++;
+    }
+  };
+
   const [baiduImgs, bingImgs, biliImgs] = await Promise.allSettled([
     fetchBaiduImages(),
     fetchBingImages(),
@@ -2303,18 +2522,26 @@ export async function searchWebImages(query, maxResults = 8) {
   const bList = baiduImgs.status === 'fulfilled' && Array.isArray(baiduImgs.value) ? baiduImgs.value : [];
   const biList = bingImgs.status === 'fulfilled' && Array.isArray(bingImgs.value) ? bingImgs.value : [];
   const blList = biliImgs.status === 'fulfilled' && Array.isArray(biliImgs.value) ? biliImgs.value : [];
+  mergeImageRound(bList, biList, blList);
 
-  let idx = 0;
-  while (results.length < capped && (idx < bList.length || idx < biList.length || idx < blList.length)) {
-    if (idx < bList.length && results.length < capped) addImage(bList[idx], 'Baidu Images');
-    if (idx < biList.length && results.length < capped) addImage(biList[idx], 'Bing Images');
-    if (idx < blList.length && results.length < capped) addImage(blList[idx], 'Bilibili Covers');
-    idx++;
+  if (inDomain && results.length < capped) {
+    const [baidu2, bing2, bili2] = await Promise.allSettled([
+      fetchBaiduImages('鞠婧祎'),
+      fetchBingImages('鞠婧祎'),
+      fetchBilibiliCovers('鞠婧祎')
+    ]);
+    mergeImageRound(
+      baidu2.status === 'fulfilled' && Array.isArray(baidu2.value) ? baidu2.value : [],
+      bing2.status === 'fulfilled' && Array.isArray(bing2.value) ? bing2.value : [],
+      bili2.status === 'fulfilled' && Array.isArray(bili2.value) ? bili2.value : []
+    );
   }
 
   if (results.length < capped) {
     try {
-      const webRes = await searchChineseWeb(query, Math.min(capped - results.length, 3));
+      const webRes = inDomain
+        ? await searchChineseWeb(query, Math.min(capped - results.length, 3))
+        : await defaultMetaSearch(query, { limit: Math.min(capped - results.length, 3) });
       const webResults = webRes?.results || [];
       const fetchedPages = await Promise.all(webResults.map(async r => {
         try {
@@ -2343,12 +2570,14 @@ export async function searchWebImages(query, maxResults = 8) {
     } catch {}
   }
 
+  const rankedResults = rerankSearchResults(results, queryPlan, { limit: capped });
   return {
     status: results.length > 0 ? 'success' : 'no_results',
-    query,
-    count: results.length,
+    query: queryPlan.originalQuery || query,
+    normalizedQuery: queryPlan.keywords,
+    count: rankedResults.length,
     sources: [...new Set(sources)],
-    results: results.slice(0, capped)
+    results: rankedResults
   };
 }
 
@@ -2628,6 +2857,21 @@ export async function searchBilibili(keyword, maxResults = 8) {
       };
     }
 
+    const extractTopicTokens = (str) => {
+      return String(str || '')
+        .replace(/(?:鞠婧祎|小鞠|jujingyi|ju\s*jing\s*yi|kiku|จวีจิ้งอี|จิ้งอี)/gi, ' ')
+        .replace(/(?:bilibili|b站|video|clip|คลิป|วิดีโอ|ค้นหา|ดู|หา|บน|ใน|ที่|มีไหม|บ้าง|ของ)/gi, ' ')
+        .trim()
+        .toLowerCase()
+        .split(/[\s,+/|&_-]+/)
+        .filter(t => t.length >= 2 || /^[0-9a-z]$/i.test(t));
+    };
+
+    const topicTokens = Array.from(new Set([
+      ...extractTopicTokens(keyword),
+      ...extractTopicTokens(effectiveKeyword)
+    ]));
+
     const items = (json.data?.result || []).slice(0, capped).map(r => {
       const bvid = r.bvid || '';
       const rawPic = r.pic || '';
@@ -2643,33 +2887,56 @@ export async function searchBilibili(keyword, maxResults = 8) {
       const rawDesc = String(r.description || '');
       const lowerDesc = rawDesc.toLowerCase();
 
-   
       const hasArtistInAuthor = lowerAuthor.includes('鞠婧祎工作室') || lowerAuthor.includes('鞠婧祎') || lowerAuthor.includes('studio');
       const hasArtistInTitle = cleanTitle.includes('鞠婧祎') || cleanTitle.includes('小鞠') || lowerTitle.includes('kiku') || lowerTitle.includes('jujingyi');
       const hasArtistInDesc = rawDesc.includes('鞠婧祎') || rawDesc.includes('小鞠');
 
       if (hasArtistInAuthor) {
-        rankScore += 100;
-      } else if (hasArtistInTitle) {
         rankScore += 80;
+      } else if (hasArtistInTitle) {
+        rankScore += 60;
       } else if (hasArtistInDesc) {
-        rankScore += 45;
+        rankScore += 25;
       } else {
-      
-        rankScore -= 40;
+        rankScore -= 50;
       }
 
+      let topicMatchTitleCount = 0;
+      let topicMatchDescCount = 0;
+      if (topicTokens.length > 0) {
+        for (const token of topicTokens) {
+          if (lowerTitle.includes(token)) topicMatchTitleCount++;
+          if (lowerDesc.includes(token)) topicMatchDescCount++;
+        }
+        if (topicMatchTitleCount > 0) {
+          rankScore += 180 + (topicMatchTitleCount * 30);
+        } else if (topicMatchDescCount > 0) {
+          rankScore += 70 + (topicMatchDescCount * 15);
+        } else {
+          rankScore -= 160;
+        }
+      }
 
-      if (lowerAuthor.includes('snh48') || lowerAuthor.includes('腾讯') || lowerAuthor.includes('优酷') || lowerAuthor.includes('爱奇艺') || lowerAuthor.includes('芒果') || lowerAuthor.includes('帕莎') || lowerAuthor.includes('prsr') || lowerAuthor.includes('官方')) {
-        rankScore += 60;
+      const hasSnhQuery = topicTokens.some(t => t.includes('snh') || t.includes('塞纳河') || t.includes('总选'));
+      if (hasSnhQuery && (lowerAuthor.includes('snh48') || lowerTitle.includes('snh48'))) {
+        rankScore += 70;
       }
 
-     
-      if (lowerTitle.includes('mv') || cleanTitle.includes('官方') || lowerTitle.includes('official') || cleanTitle.includes('舞台') || cleanTitle.includes('现场') || cleanTitle.includes('代言') || cleanTitle.includes('花絮') || cleanTitle.includes('广告') || lowerTitle.includes('prsr') || cleanTitle.includes('帕莎')) {
-        rankScore += 35;
+      const hasPrsrQuery = topicTokens.some(t => t.includes('prsr') || t.includes('帕莎') || t.includes('pasha') || t.includes('眼镜'));
+      if (hasPrsrQuery && (lowerAuthor.includes('帕莎') || lowerAuthor.includes('prsr') || lowerTitle.includes('帕莎') || lowerTitle.includes('prsr'))) {
+        rankScore += 80;
       }
 
-      if (play > 0) rankScore += Math.min(Math.log10(play) * 10, 50);
+      const isVerifiedMediaOrStudio = lowerAuthor.includes('鞠婧祎工作室') || lowerAuthor.includes('腾讯') || lowerAuthor.includes('优酷') || lowerAuthor.includes('爱奇艺') || lowerAuthor.includes('芒果') || lowerAuthor.includes('官方');
+      if (isVerifiedMediaOrStudio) {
+        rankScore += 30;
+      }
+
+      if (lowerTitle.includes('mv') || cleanTitle.includes('官方') || lowerTitle.includes('official') || cleanTitle.includes('舞台') || cleanTitle.includes('现场') || cleanTitle.includes('代言') || cleanTitle.includes('花絮') || cleanTitle.includes('广告')) {
+        rankScore += 25;
+      }
+
+      if (play > 0) rankScore += Math.min(Math.log10(play) * 8, 30);
 
       return {
         title: cleanTitle,
@@ -2683,19 +2950,31 @@ export async function searchBilibili(keyword, maxResults = 8) {
         embedUrl: bvid ? buildBilibiliEmbedUrl(bvid) : '',
         description: rawDesc.slice(0, 300),
         url: bvid ? `https://www.bilibili.com/video/${bvid}` : '',
-        rankScore: Math.round(rankScore * 10) / 10
+        rankScore: Math.round(rankScore * 10) / 10,
+        topicMatchTitleCount,
+        topicMatchDescCount
       };
     }).filter(r => r.title && r.url);
 
-    items.sort((a, b) => (b.rankScore || 0) - (a.rankScore || 0));
+    const hasTopicSpecificTokens = topicTokens.length > 0;
+    const hasAnyTopicMatch = items.some(item => item.topicMatchTitleCount > 0 || item.topicMatchDescCount > 0);
+
+    const finalItems = items.filter(r => {
+      if (hasTopicSpecificTokens && hasAnyTopicMatch) {
+        return (r.topicMatchTitleCount > 0 || r.topicMatchDescCount > 0) && r.rankScore > 0;
+      }
+      return r.rankScore > -50;
+    });
+
+    finalItems.sort((a, b) => (b.rankScore || 0) - (a.rankScore || 0));
 
     return {
-      status: items.length > 0 ? 'success' : 'no_results',
+      status: finalItems.length > 0 ? 'success' : 'no_results',
       keyword,
       searchQuery: effectiveKeyword,
-      count: items.length,
+      count: finalItems.length,
       source: 'Bilibili',
-      results: items
+      results: finalItems
     };
   } catch (err) {
     return {