geo/apps/desktop-client/src/main/adapters/wenxin.ts

import type { JsonValue, MonitoringSourceItem } from '@geo/shared-types'
import type { Session } from 'electron/main'
import type { Page as PlaywrightPage } from 'playwright-core'

import type { MonitorAdapter } from './base'
import { normalizeText, sessionCookieFetchJson } from './common'

const WENXIN_BOOTSTRAP_URL = 'https://yiyan.baidu.com/'
const WENXIN_HISTORY_URL = 'https://yiyan.baidu.com/eb/chat/history'
const WENXIN_PAGE_READY_TIMEOUT_MS = 20_000
const WENXIN_QUERY_TIMEOUT_MS = 180_000
const WENXIN_QUERY_POLL_INTERVAL_MS = 1_250
const WENXIN_STABLE_POLLS_REQUIRED = 3
const WENXIN_STREAM_CAPTURE_LIMIT = 8
const WENXIN_STREAM_CAPTURE_BODY_LIMIT = 800_000
const WENXIN_EDITOR_SELECTOR =
  'div[role="textbox"].editable__T7WAW4uW, div[role="textbox"], [contenteditable="true"]'
const WENXIN_LOGIN_SIGNALS = ['未登录', '请先登录', '登录', '扫码登录', '百度账号登录']
const WENXIN_RISK_CONTROL_PATTERN =
  /(当前访问环境存在异常|访问环境存在异常|环境存在异常|更换浏览器再尝试提问|当前环境异常)/i
const WENXIN_LOGIN_REQUIRED_PATTERN = /(请先登录|未登录|登录后|登录态失效|重新登录)/i
const WENXIN_PLACEHOLDER_PATTERN =
  /^(?:正在生成中(?:\.\.\.)?|思考中(?:\.\.\.)?|内容由AI生成，仅供参考，请仔细甄别|全选)$/
const WENXIN_TABLE_HTML_PATTERN = /<table\b[\s\S]*?<\/table>/gi
const WENXIN_REFERENCE_TRIGGER_PATTERN =
  /(参考\s*\d+\s*(?:个网页|条网页信息源)|found\s*\d+\s*web pages)/i
const WENXIN_CITATION_FIELD_PATTERN =
  /^(?:citations?|citationInfo|pluginCitations?|pluginCitation|reference(?:s|List|_list|Cards|_cards|Docs|_docs|Info|_info)?|source(?:s|List|_list|Cards|_cards|Info|_info)?|web(?:References?|Sources?|Pages?|Docs?|Results?)|page(?:References?|Sources?))$/i
const WENXIN_SEARCH_FIELD_PATTERN =
  /^(?:searchCitations?|search_results?|searchResultList|search_result_list|webListData|webList|web(?:References?|Sources?|Pages?|Docs?|Results?)|browseReferences?|browse_results?|grounding(?:_results?)?)$/i
const WENXIN_REDIRECT_QUERY_KEYS = [
  'url',
  'u',
  'target',
  'targetUrl',
  'target_url',
  'link',
  'href',
  'sourceUrl',
  'source_url',
  'originUrl',
  'origin_url',
  'realUrl',
  'real_url',
  'redirect',
  'redirect_url',
  'jump',
  'jump_url',
  'to',
]

type WenxinPageSourceLink = {
  url: string
  title: string | null
  text: string | null
  siteName: string | null
}

type WenxinPageSnapshot = {
  url: string
  title: string | null
  loginRequired: boolean
  loginReason: string | null
  isLogin: boolean | null
  modelLabel: string | null
  userName: string | null
  editorText: string | null
  bodySample: string | null
  referenceTriggerText: string | null
  referenceLinks: WenxinPageSourceLink[]
}

type WenxinCapturedStreamRecord = {
  id: string
  url: string
  method: string
  status: number | null
  contentType: string | null
  requestBody: string | null
  body: string
  done: boolean
  error: string | null
  startedAt: number
  updatedAt: number
}

type WenxinStreamSummary = {
  sessionId: string | null
  userChatId: string | null
  botChatId: string | null
  providerModel: string | null
  providerRequestID: string | null
  answer: string | null
  withdrawText: string | null
  errorMessage: string | null
  isBanned: boolean
  sessionBan: boolean
  sessionDel: boolean
  isEnded: boolean
  eventCount: number
  captureDone: boolean
  captureError: string | null
}

type WenxinHistorySummary = {
  sessionId: string | null
  loginRequired: boolean
  providerModel: string | null
  providerRequestID: string | null
  latestChatId: string | null
  latestChatMode: string | null
  latestChatStop: number | boolean | null
  state: number | string | null
  answer: string | null
  answerBlocks: string[]
  citations: MonitoringSourceItem[]
  searchResults: MonitoringSourceItem[]
  withdrawText: string | null
  errorMessage: string | null
  generating: boolean
  signature: string | null
}

type WenxinObservation = {
  sessionId: string | null
  providerModel: string | null
  providerRequestID: string | null
  answer: string | null
  answerBlocks: string[]
  citations: MonitoringSourceItem[]
  searchResults: MonitoringSourceItem[]
  withdrawText: string | null
  errorMessage: string | null
  historyGenerating: boolean
  streamDone: boolean
  signature: string | null
}

type WenxinWaitResult =
  | {
      ok: true
      stream: WenxinStreamSummary
      history: WenxinHistorySummary | null
      observation: WenxinObservation
      stablePollCount: number
      elapsedMs: number
    }
  | {
      ok: false
      error: string
      detail: string | null
      stream: WenxinStreamSummary
      history: WenxinHistorySummary | null
      observation: WenxinObservation
      stablePollCount: number
      elapsedMs: number
    }

function isRecord(value: unknown): value is Record<string, unknown> {
  return typeof value === 'object' && value !== null && !Array.isArray(value)
}

function normalizeOptionalString(value: unknown): string | null {
  if (typeof value !== 'string') {
    return null
  }

  const trimmed = value.trim()
  return trimmed ? trimmed : null
}

function normalizeBlockText(value: string | null): string | null {
  const normalized = normalizeOptionalString(value)
  if (!normalized) {
    return null
  }

  return (
    normalized
      .replace(/\u00a0/g, ' ')
      .replace(/\r/g, '')
      .replace(/\n{3,}/g, '\n\n')
      .trim() || null
  )
}

function decodeUrlCandidate(value: string): string {
  let decoded = value.trim()
  for (let index = 0; index < 3; index += 1) {
    try {
      const next = decodeURIComponent(decoded)
      if (next === decoded) {
        break
      }
      decoded = next
    } catch {
      break
    }
  }
  return decoded
}

function coercePotentialUrl(value: string): string {
  const trimmed = decodeUrlCandidate(value)
  if (/^\/\//.test(trimmed)) {
    return `https:${trimmed}`
  }
  if (/^[a-z0-9-]+(?:\.[a-z0-9-]+)+(?:[/?#].*)?$/i.test(trimmed)) {
    return `https://${trimmed}`
  }
  return trimmed
}

function normalizeAbsoluteUrl(value: string): string | null {
  try {
    const url = new URL(coercePotentialUrl(value))
    url.hash = ''
    return url.toString()
  } catch {
    return null
  }
}

function extractRedirectTargetUrl(value: string): string | null {
  const direct = normalizeAbsoluteUrl(value)
  if (!direct) {
    return null
  }

  try {
    const url = new URL(direct)
    for (const key of WENXIN_REDIRECT_QUERY_KEYS) {
      const raw = url.searchParams.get(key)
      if (!raw) {
        continue
      }
      const normalized = normalizeAbsoluteUrl(raw)
      if (normalized) {
        return normalized
      }
    }
  } catch {
    return null
  }

  const encodedMatch = direct.match(/https?%3A%2F%2F[^&\s"'<>]+/i)
  if (encodedMatch) {
    return normalizeAbsoluteUrl(encodedMatch[0])
  }

  const plainMatch = direct.match(/https?:\/\/[^&\s"'<>]+/i)
  if (plainMatch && plainMatch[0] !== direct) {
    return normalizeAbsoluteUrl(plainMatch[0])
  }

  return null
}

function normalizeUrl(value: unknown): string | null {
  const input = normalizeText(value)
  if (!input) {
    return null
  }

  return extractRedirectTargetUrl(input) ?? normalizeAbsoluteUrl(input)
}

function isLikelyWenxinSourceUrl(url: string | null): boolean {
  if (!url) {
    return false
  }

  try {
    const parsed = new URL(url)
    const host = parsed.hostname.toLowerCase()
    if (!/^https?:$/i.test(parsed.protocol)) {
      return false
    }
    if (
      host === 'yiyan.baidu.com' ||
      host.endsWith('.yiyan.baidu.com') ||
      host === 'eb-static.cdn.bcebos.com' ||
      host === 'ppui-static-wap.cdn.bcebos.com'
    ) {
      return false
    }
    return true
  } catch {
    return false
  }
}

function readStringByKeys(source: unknown, keys: string[]): string | null {
  if (!isRecord(source)) {
    return null
  }

  for (const key of keys) {
    const value = normalizeOptionalString(source[key])
    if (value) {
      return value
    }
  }

  return null
}

function readBooleanByKeys(source: unknown, keys: string[]): boolean | null {
  if (!isRecord(source)) {
    return null
  }

  for (const key of keys) {
    const value = source[key]
    if (typeof value === 'boolean') {
      return value
    }
  }

  return null
}

function readNumberByKeys(source: unknown, keys: string[]): number | null {
  if (!isRecord(source)) {
    return null
  }

  for (const key of keys) {
    const value = source[key]
    if (typeof value === 'number' && Number.isFinite(value)) {
      return value
    }
  }

  return null
}

function normalizeStopFlag(value: unknown): number | boolean | null {
  if (typeof value === 'boolean') {
    return value
  }

  if (typeof value === 'number' && Number.isFinite(value)) {
    return value
  }

  if (typeof value === 'string') {
    const trimmed = value.trim().toLowerCase()
    if (!trimmed) {
      return null
    }
    if (trimmed === 'true') {
      return true
    }
    if (trimmed === 'false') {
      return false
    }

    const numericValue = Number(trimmed)
    if (Number.isFinite(numericValue)) {
      return numericValue
    }
  }

  return null
}

function mergeText(current: string | null, next: string | null): string | null {
  const left = normalizeBlockText(current)
  const right = normalizeBlockText(next)
  if (!left) {
    return right
  }
  if (!right) {
    return left
  }

  if (right.startsWith(left)) {
    return right
  }
  if (left.startsWith(right)) {
    return left
  }
  if (left.endsWith(right)) {
    return left
  }
  if (right.endsWith(left)) {
    return right
  }

  return `${left}${right}`
}

function safeParseJSON(value: string): unknown {
  try {
    return JSON.parse(value)
  } catch {
    return null
  }
}

function dedupeSourceItems(items: MonitoringSourceItem[]): MonitoringSourceItem[] {
  const keyed = new Map<string, MonitoringSourceItem>()
  for (const item of items) {
    const key = normalizeUrl(item.normalized_url ?? item.url)
    if (!key) {
      continue
    }

    const existing = keyed.get(key)
    if (!existing) {
      keyed.set(key, {
        ...item,
        normalized_url: key,
      })
      continue
    }

    keyed.set(key, {
      ...existing,
      title: existing.title ?? item.title ?? null,
      site_name: existing.site_name ?? item.site_name ?? null,
      host: existing.host ?? item.host ?? null,
    })
  }

  return Array.from(keyed.values())
}

function buildSourceItem(input: unknown): MonitoringSourceItem | null {
  if (!isRecord(input)) {
    return null
  }

  const url = normalizeUrl(
    input.url ??
      input.href ??
      input.link ??
      input.pageUrl ??
      input.page_url ??
      input.webUrl ??
      input.web_url ??
      input.realUrl ??
      input.real_url ??
      input.displayUrl ??
      input.display_url ??
      input.originUrl ??
      input.origin_url ??
      input.targetUrl ??
      input.target_url ??
      input.sourceUrl ??
      input.source_url,
  )
  if (!url || !isLikelyWenxinSourceUrl(url)) {
    return null
  }

  let host: string | null = null
  try {
    host = new URL(url).hostname || null
  } catch {
    host = null
  }

  return {
    url,
    title: normalizeText(input.title ?? input.name ?? input.text ?? input.label ?? input.desc),
    site_name: normalizeText(
      input.site_name ?? input.siteName ?? input.site ?? input.domain ?? input.source,
    ),
    normalized_url: url,
    host,
  }
}

function collectSourceBucketByFieldPattern(
  input: unknown,
  bucket: MonitoringSourceItem[],
  pattern: RegExp,
  depth = 0,
): void {
  if (depth > 10 || input == null) {
    return
  }

  if (Array.isArray(input)) {
    for (const item of input) {
      collectSourceBucketByFieldPattern(item, bucket, pattern, depth + 1)
    }
    return
  }

  if (!isRecord(input)) {
    return
  }

  for (const [key, value] of Object.entries(input)) {
    if (pattern.test(key)) {
      collectSourceBucket(value, bucket, depth + 1)
    }

    if (Array.isArray(value) || isRecord(value)) {
      collectSourceBucketByFieldPattern(value, bucket, pattern, depth + 1)
    }
  }
}

function collectSourceBucket(input: unknown, bucket: MonitoringSourceItem[], depth = 0): void {
  if (depth > 10 || input == null) {
    return
  }

  if (Array.isArray(input)) {
    for (const item of input) {
      collectSourceBucket(item, bucket, depth + 1)
    }
    return
  }

  const sourceItem = buildSourceItem(input)
  if (sourceItem) {
    bucket.push(sourceItem)
  }

  if (!isRecord(input)) {
    return
  }

  for (const value of Object.values(input)) {
    if (Array.isArray(value) || isRecord(value)) {
      collectSourceBucket(value, bucket, depth + 1)
    }
  }
}

function decodeHtmlEntities(value: string): string {
  const namedEntityMap: Record<string, string> = {
    amp: '&',
    lt: '<',
    gt: '>',
    quot: '"',
    apos: "'",
    nbsp: ' ',
    mdash: '-',
    ndash: '-',
  }

  return value.replace(/&(#x?[0-9a-f]+|[a-z]+);/gi, (matched, entity: string) => {
    const lower = entity.toLowerCase()
    if (lower in namedEntityMap) {
      return namedEntityMap[lower]
    }

    if (lower.startsWith('#x')) {
      const codePoint = Number.parseInt(lower.slice(2), 16)
      return Number.isFinite(codePoint) ? String.fromCodePoint(codePoint) : matched
    }

    if (lower.startsWith('#')) {
      const codePoint = Number.parseInt(lower.slice(1), 10)
      return Number.isFinite(codePoint) ? String.fromCodePoint(codePoint) : matched
    }

    return matched
  })
}

function stripHtml(value: string): string {
  return decodeHtmlEntities(
    value
      .replace(/<br\s*\/?>/gi, '\n')
      .replace(/<\/(?:p|div|section|article|li|ul|ol|h[1-6])>/gi, '\n')
      .replace(/<li\b[^>]*>/gi, '- ')
      .replace(/<[^>]+>/g, ' '),
  )
    .replace(/[ \t]+\n/g, '\n')
    .replace(/\n[ \t]+/g, '\n')
    .replace(/[ \t]{2,}/g, ' ')
}

function htmlTableToMarkdown(tableHtml: string): string | null {
  const rowMatches = [...tableHtml.matchAll(/<tr\b[^>]*>([\s\S]*?)<\/tr>/gi)]
  if (!rowMatches.length) {
    return null
  }

  const rows = rowMatches
    .map((match) => [...match[1].matchAll(/<t[hd]\b[^>]*>([\s\S]*?)<\/t[hd]>/gi)])
    .map((cells) =>
      cells
        .map((cell) => normalizeBlockText(stripHtml(cell[1] ?? '')) ?? '')
        .map((cell) => cell.replace(/\|/g, '\\|').replace(/\n+/g, '<br/>')),
    )
    .filter((cells) => cells.some((cell) => cell.length > 0))

  if (!rows.length) {
    return null
  }

  const header = rows[0]
  const body = rows.slice(1)
  const separator = header.map(() => '---')
  const lines = [
    `| ${header.join(' | ')} |`,
    `| ${separator.join(' | ')} |`,
    ...body.map((row) => `| ${row.join(' | ')} |`),
  ]
  return normalizeBlockText(lines.join('\n'))
}

function textSegmentsFromHtml(value: string): string[] {
  const result: string[] = []
  const push = (segment: string | null) => {
    const normalized = normalizeBlockText(segment)
    if (normalized) {
      result.push(normalized)
    }
  }

  const tables = [...value.matchAll(WENXIN_TABLE_HTML_PATTERN)]
  for (const match of tables) {
    push(htmlTableToMarkdown(match[0]))
  }

  const withoutTables = value.replace(WENXIN_TABLE_HTML_PATTERN, '\n')
  push(stripHtml(withoutTables))
  return result
}

function extractTextSegments(value: unknown, result: string[], seen: Set<string>, depth = 0): void {
  if (depth > 6 || value == null) {
    return
  }

  const push = (segment: string | null) => {
    const normalized = normalizeBlockText(segment)
    if (!normalized || WENXIN_PLACEHOLDER_PATTERN.test(normalized)) {
      return
    }

    const key = normalized.replace(/\s+/g, ' ')
    if (seen.has(key)) {
      return
    }
    seen.add(key)
    result.push(normalized)
  }

  if (typeof value === 'string') {
    if (/<[a-z][\s\S]*>/i.test(value)) {
      for (const segment of textSegmentsFromHtml(value)) {
        push(segment)
      }
      return
    }
    push(value)
    return
  }

  if (Array.isArray(value)) {
    for (const item of value) {
      extractTextSegments(item, result, seen, depth + 1)
    }
    return
  }

  if (!isRecord(value)) {
    return
  }

  for (const key of ['content', 'text', 'tips', 'quoteText', 'markdown', 'html']) {
    if (key in value) {
      extractTextSegments(value[key], result, seen, depth + 1)
    }
  }

  for (const key of ['htmlMetaInfo', 'metaData', 'resInfo']) {
    if (key in value) {
      extractTextSegments(value[key], result, seen, depth + 1)
    }
  }
}

function extractWenxinMessageBlocks(messages: unknown): string[] {
  const result: string[] = []
  const seen = new Set<string>()

  if (Array.isArray(messages)) {
    for (const message of messages) {
      extractTextSegments(message, result, seen)
    }
    return result
  }

  extractTextSegments(messages, result, seen)
  return result
}

function answerQualityScore(value: string | null): number {
  const normalized = normalizeBlockText(value)
  if (!normalized) {
    return Number.NEGATIVE_INFINITY
  }

  let score = normalized.length
  if (/\| .+ \|/.test(normalized)) {
    score += 120
  }
  if (/(?:^|\n)(?:[-*+]\s+|\d+\.\s+|#{1,6}\s+)/m.test(normalized)) {
    score += 40
  }
  if (WENXIN_PLACEHOLDER_PATTERN.test(normalized)) {
    score -= 500
  }
  return score
}

function selectBestAnswer(primary: string | null, secondary: string | null): string | null {
  const first = normalizeBlockText(primary)
  const second = normalizeBlockText(secondary)
  if (!first) {
    return second
  }
  if (!second) {
    return first
  }

  return answerQualityScore(second) >= answerQualityScore(first) ? second : first
}

function extractQuestionText(payload: Record<string, unknown>): string {
  const candidates = [
    payload.question_text,
    payload.questionText,
    payload.query,
    payload.question,
    payload.prompt,
    payload.content,
    payload.title,
  ]

  for (const candidate of candidates) {
    const text = normalizeOptionalString(candidate)
    if (text) {
      return text
    }
  }

  throw new Error('wenxin_question_text_missing')
}

function toJsonValue(value: unknown, depth = 0): JsonValue {
  if (value == null || depth > 12) {
    return null
  }

  if (typeof value === 'string' || typeof value === 'boolean') {
    return value
  }

  if (typeof value === 'number') {
    return Number.isFinite(value) ? value : null
  }

  if (Array.isArray(value)) {
    return value.map((item) => toJsonValue(item, depth + 1))
  }

  if (!isRecord(value)) {
    return null
  }

  const result: Record<string, JsonValue> = {}
  for (const [key, item] of Object.entries(value)) {
    result[key] = toJsonValue(item, depth + 1)
  }
  return result
}

function toJsonSources(items: MonitoringSourceItem[]): JsonValue[] {
  return items.map((item) => ({
    url: item.url,
    title: item.title ?? null,
    site_name: item.site_name ?? null,
    site_key: item.site_key ?? null,
    normalized_url: item.normalized_url ?? null,
    host: item.host ?? null,
    registrable_domain: item.registrable_domain ?? null,
    subdomain: item.subdomain ?? null,
    suffix: item.suffix ?? null,
    article_id: item.article_id ?? null,
    publish_record_id: item.publish_record_id ?? null,
    resolution_status: item.resolution_status ?? null,
    resolution_confidence: item.resolution_confidence ?? null,
  }))
}

function buildAdapterError(
  code: string,
  message: string,
  extras: Record<string, JsonValue> = {},
): Record<string, JsonValue> {
  return {
    code,
    message,
    ...extras,
  }
}

function emptyWenxinStreamSummary(): WenxinStreamSummary {
  return {
    sessionId: null,
    userChatId: null,
    botChatId: null,
    providerModel: null,
    providerRequestID: null,
    answer: null,
    withdrawText: null,
    errorMessage: null,
    isBanned: false,
    sessionBan: false,
    sessionDel: false,
    isEnded: false,
    eventCount: 0,
    captureDone: false,
    captureError: null,
  }
}

function parseSSERecord(record: string): { event: string; payload: unknown } | null {
  const lines = record.split(/\r?\n/)
  let eventName = 'message'
  const dataLines: string[] = []

  for (const line of lines) {
    if (!line || line.startsWith(':') || line.startsWith('ERR:')) {
      continue
    }
    if (line.startsWith('event:')) {
      eventName = line.slice(6).trim() || eventName
      continue
    }
    if (line.startsWith('data:')) {
      dataLines.push(line.slice(5).trimStart())
    }
  }

  if (!dataLines.length) {
    return null
  }

  return {
    event: eventName,
    payload: safeParseJSON(dataLines.join('\n')),
  }
}

function parseWenxinSSEBody(
  body: string,
  captureDone = false,
  captureError: string | null = null,
): WenxinStreamSummary {
  const summary = emptyWenxinStreamSummary()
  summary.captureDone = captureDone
  summary.captureError = captureError

  if (!body) {
    return summary
  }

  const records = body
    .split(/\r?\n\r?\n+/)
    .map((record) => record.trim())
    .filter(Boolean)

  for (const record of records) {
    const parsed = parseSSERecord(record)
    if (!parsed || !isRecord(parsed.payload)) {
      continue
    }

    summary.eventCount += 1

    if (parsed.event === 'major') {
      const payloadData = isRecord(parsed.payload.data) ? parsed.payload.data : null
      const createChat = isRecord(payloadData?.createChatResponseVoCommonResult)
        ? payloadData.createChatResponseVoCommonResult
        : null
      const createChatData = isRecord(createChat?.data) ? createChat.data : null
      const createSession = isRecord(payloadData?.createSessionResponseVoCommonResult)
        ? payloadData.createSessionResponseVoCommonResult
        : null
      const createSessionData = isRecord(createSession?.data) ? createSession.data : null
      const botChat = isRecord(createChatData?.botChat) ? createChatData.botChat : null
      const userChat = isRecord(createChatData?.chat) ? createChatData.chat : null

      summary.sessionId = readStringByKeys(createSessionData, ['sessionId']) ?? summary.sessionId
      summary.userChatId = readStringByKeys(userChat, ['id']) ?? summary.userChatId
      summary.botChatId = readStringByKeys(botChat, ['id']) ?? summary.botChatId
      summary.providerModel =
        readStringByKeys(createSessionData, ['model']) ??
        readStringByKeys(botChat, ['modelSign', 'model']) ??
        summary.providerModel
      summary.providerRequestID =
        readStringByKeys(botChat, ['messageId']) ??
        readStringByKeys(createChat, ['logId']) ??
        readStringByKeys(parsed.payload, ['logId']) ??
        summary.providerRequestID
      summary.withdrawText =
        readStringByKeys(createChatData, ['withdrawText']) ?? summary.withdrawText
      summary.errorMessage =
        readStringByKeys(createChatData, ['errMessage']) ??
        readStringByKeys(createChat, ['msg']) ??
        summary.errorMessage
      summary.isBanned = readBooleanByKeys(createChatData, ['isBanned']) ?? summary.isBanned
      summary.sessionBan = readBooleanByKeys(createChatData, ['sessionBan']) ?? summary.sessionBan
      summary.sessionDel = readBooleanByKeys(createChatData, ['sessionDel']) ?? summary.sessionDel

      const initialAnswer = extractWenxinMessageBlocks(botChat?.message).join('\n\n')
      summary.answer = selectBestAnswer(summary.answer, initialAnswer)
      continue
    }

    if (parsed.event === 'message') {
      const payloadData = isRecord(parsed.payload.data) ? parsed.payload.data : null
      const chatID = readStringByKeys(payloadData, ['chat_id', 'chatId'])
      if (summary.botChatId && chatID && summary.botChatId !== chatID) {
        continue
      }
      if (!summary.botChatId && chatID) {
        summary.botChatId = chatID
      }

      const tokensAll = readStringByKeys(payloadData, ['tokens_all'])
      const delta = readStringByKeys(payloadData, ['content', 'text'])
      summary.answer = tokensAll
        ? selectBestAnswer(summary.answer, tokensAll)
        : mergeText(summary.answer, delta)
      summary.isEnded = readBooleanByKeys(payloadData, ['is_end']) ?? summary.isEnded

      const messageText =
        readStringByKeys(parsed.payload, ['msg']) ??
        readStringByKeys(payloadData, ['msg', 'message', 'detail'])
      const code = readNumberByKeys(parsed.payload, ['code'])
      if (code !== null && code !== 0 && messageText) {
        summary.errorMessage = messageText
      }
    }
  }

  if (!summary.errorMessage && captureError && !/AbortError/i.test(captureError)) {
    summary.errorMessage = captureError
  }

  return summary
}

function findLatestRobotChat(chats: unknown[]): Record<string, unknown> | null {
  for (let index = chats.length - 1; index >= 0; index -= 1) {
    const item = chats[index]
    if (!isRecord(item)) {
      continue
    }

    const role = readStringByKeys(item, ['role'])
    if (role !== 'robot') {
      continue
    }

    const parent = readStringByKeys(item, ['parent'])
    if (parent === '0' && index < chats.length - 1) {
      continue
    }

    return item
  }

  return null
}

function parseWenxinHistoryResponse(
  payload: unknown,
  sessionId: string | null = null,
): WenxinHistorySummary {
  const code = readNumberByKeys(payload, ['code'])
  const message = readStringByKeys(payload, ['msg'])
  if (code === 1001 || WENXIN_LOGIN_REQUIRED_PATTERN.test(message ?? '')) {
    return {
      sessionId,
      loginRequired: true,
      providerModel: null,
      providerRequestID: null,
      latestChatId: null,
      latestChatMode: null,
      latestChatStop: null,
      state: null,
      answer: null,
      answerBlocks: [],
      citations: [],
      searchResults: [],
      withdrawText: null,
      errorMessage: message ?? '请先登录',
      generating: false,
      signature: null,
    }
  }

  const data = isRecord(payload) && isRecord(payload.data) ? payload.data : null
  const chats = Array.isArray(data?.chats) ? data.chats : []
  const latestChat = findLatestRobotChat(chats)
  const answerBlocks = extractWenxinMessageBlocks(latestChat?.message)
  const answer = normalizeBlockText(answerBlocks.join('\n\n'))

  const citationBucket: MonitoringSourceItem[] = []
  collectSourceBucketByFieldPattern(latestChat, citationBucket, WENXIN_CITATION_FIELD_PATTERN)
  if (!citationBucket.length) {
    collectSourceBucketByFieldPattern(data, citationBucket, WENXIN_CITATION_FIELD_PATTERN)
  }

  const searchBucket: MonitoringSourceItem[] = []
  collectSourceBucketByFieldPattern(latestChat, searchBucket, WENXIN_SEARCH_FIELD_PATTERN)
  if (!searchBucket.length) {
    collectSourceBucketByFieldPattern(data, searchBucket, WENXIN_SEARCH_FIELD_PATTERN)
  }

  const citations = dedupeSourceItems([...citationBucket, ...searchBucket])
  const searchResults = dedupeSourceItems(searchBucket)
  const withdrawText =
    readStringByKeys(latestChat, ['withdrawText']) ?? readStringByKeys(data, ['withdrawText'])
  const errorMessage =
    readStringByKeys(latestChat, ['errMessage']) ?? (code !== null && code !== 0 ? message : null)
  const latestChatStop = normalizeStopFlag(latestChat?.stop)
  const latestChatMode = readStringByKeys(latestChat, ['mode'])
  const state = data?.state ?? null
  const generating =
    latestChatStop === 0 ||
    (answer !== null && WENXIN_PLACEHOLDER_PATTERN.test(answer)) ||
    (!answer &&
      !withdrawText &&
      !errorMessage &&
      !citations.length &&
      !searchResults.length &&
      Boolean(latestChat))

  return {
    sessionId,
    loginRequired: false,
    providerModel: readStringByKeys(latestChat, ['modelSign', 'model']),
    providerRequestID:
      readStringByKeys(latestChat, ['messageId']) ?? readStringByKeys(payload, ['logId']),
    latestChatId: readStringByKeys(latestChat, ['id']),
    latestChatMode,
    latestChatStop,
    state: typeof state === 'number' || typeof state === 'string' ? state : null,
    answer,
    answerBlocks,
    citations,
    searchResults,
    withdrawText,
    errorMessage,
    generating,
    signature: JSON.stringify({
      latestChatId: readStringByKeys(latestChat, ['id']),
      latestChatMode,
      latestChatStop,
      state: typeof state === 'number' || typeof state === 'string' ? state : null,
      answer,
      citationCount: citations.length,
      searchCount: searchResults.length,
      withdrawText,
      errorMessage,
    }),
  }
}

function mergeWenxinObservation(
  stream: WenxinStreamSummary,
  history: WenxinHistorySummary | null,
): WenxinObservation {
  const citations = dedupeSourceItems([...(history?.citations ?? [])])
  const searchResults = dedupeSourceItems([...(history?.searchResults ?? [])])
  const answer = selectBestAnswer(stream.answer, history?.answer ?? null)
  const answerBlocks = history?.answerBlocks.length ? history.answerBlocks : answer ? [answer] : []
  const withdrawText = history?.withdrawText ?? stream.withdrawText
  const errorMessage = history?.errorMessage ?? stream.errorMessage ?? stream.captureError

  return {
    sessionId: history?.sessionId ?? stream.sessionId,
    providerModel: history?.providerModel ?? stream.providerModel,
    providerRequestID: history?.providerRequestID ?? stream.providerRequestID,
    answer,
    answerBlocks,
    citations,
    searchResults,
    withdrawText,
    errorMessage,
    historyGenerating: history?.generating ?? false,
    streamDone: stream.captureDone || stream.isEnded,
    signature: JSON.stringify({
      sessionId: history?.sessionId ?? stream.sessionId,
      answer,
      answerBlockCount: answerBlocks.length,
      citationCount: citations.length,
      searchCount: searchResults.length,
      withdrawText,
      errorMessage,
      historyGenerating: history?.generating ?? false,
      streamDone: stream.captureDone || stream.isEnded,
    }),
  }
}

function isWenxinRiskControlMessage(value: string | null | undefined): boolean {
  return WENXIN_RISK_CONTROL_PATTERN.test(value ?? '')
}

function isWenxinLoginMessage(value: string | null | undefined): boolean {
  return WENXIN_LOGIN_REQUIRED_PATTERN.test(value ?? '')
}

function isObservationComplete(observation: WenxinObservation): boolean {
  if (observation.historyGenerating) {
    return false
  }

  if (observation.answer && WENXIN_PLACEHOLDER_PATTERN.test(observation.answer)) {
    return false
  }

  return Boolean(
    observation.answer || observation.citations.length || observation.searchResults.length,
  )
}

async function sleep(ms: number, signal?: AbortSignal): Promise<void> {
  await new Promise<void>((resolve, reject) => {
    if (signal?.aborted) {
      reject(new Error('adapter_aborted'))
      return
    }

    const timer = setTimeout(() => {
      signal?.removeEventListener('abort', onAbort)
      resolve()
    }, ms)

    const onAbort = () => {
      clearTimeout(timer)
      signal?.removeEventListener('abort', onAbort)
      reject(new Error('adapter_aborted'))
    }

    signal?.addEventListener('abort', onAbort, { once: true })
  })
}

async function ensurePageOnWenxinChat(page: PlaywrightPage, signal: AbortSignal): Promise<void> {
  if (signal.aborted) {
    throw new Error('adapter_aborted')
  }

  await page.goto(WENXIN_BOOTSTRAP_URL, {
    waitUntil: 'domcontentloaded',
    timeout: WENXIN_PAGE_READY_TIMEOUT_MS,
  })

  await page
    .waitForLoadState('domcontentloaded', {
      timeout: WENXIN_PAGE_READY_TIMEOUT_MS,
    })
    .catch(() => undefined)
  await page
    .waitForLoadState('networkidle', {
      timeout: 3_000,
    })
    .catch(() => undefined)
}

async function readWenxinPageSnapshot(page: PlaywrightPage): Promise<WenxinPageSnapshot> {
  return await page.evaluate(
    ({
      editorSelector,
      loginSignals,
      triggerPatternSource,
      citationFieldPatternSource,
      searchFieldPatternSource,
      redirectQueryKeys,
    }) => {
      const isObjectRecord = (value: unknown): value is Record<string, unknown> =>
        typeof value === 'object' && value !== null && !Array.isArray(value)

      const normalize = (value: unknown): string | null => {
        if (typeof value !== 'string') {
          return null
        }
        const trimmed = value
          .replace(/\u00a0/g, ' ')
          .replace(/\r/g, '')
          .replace(/\n{3,}/g, '\n\n')
          .trim()
        return trimmed || null
      }

      const decodeUrlCandidate = (value: string): string => {
        let decoded = value.trim()
        for (let index = 0; index < 3; index += 1) {
          try {
            const next = decodeURIComponent(decoded)
            if (next === decoded) {
              break
            }
            decoded = next
          } catch {
            break
          }
        }
        return decoded
      }

      const coercePotentialUrl = (value: string): string => {
        const trimmed = decodeUrlCandidate(value)
        if (/^\/\//.test(trimmed)) {
          return `https:${trimmed}`
        }
        if (/^[a-z0-9-]+(?:\.[a-z0-9-]+)+(?:[/?#].*)?$/i.test(trimmed)) {
          return `https://${trimmed}`
        }
        return trimmed
      }

      const normalizeAbsoluteHref = (value: string): string | null => {
        try {
          const url = new URL(coercePotentialUrl(value), window.location.href)
          url.hash = ''
          return url.toString()
        } catch {
          return null
        }
      }

      const extractRedirectTargetHref = (value: string): string | null => {
        const direct = normalizeAbsoluteHref(value)
        if (!direct) {
          return null
        }

        try {
          const url = new URL(direct)
          for (const key of redirectQueryKeys) {
            const raw = url.searchParams.get(key)
            if (!raw) {
              continue
            }
            const normalized = normalizeAbsoluteHref(raw)
            if (normalized) {
              return normalized
            }
          }
        } catch {
          return null
        }

        const encodedMatch = direct.match(/https?%3A%2F%2F[^&\s"'<>]+/i)
        if (encodedMatch) {
          return normalizeAbsoluteHref(encodedMatch[0])
        }

        return null
      }

      const normalizeHref = (value: unknown): string | null => {
        if (typeof value !== 'string') {
          return null
        }

        const trimmed = value.trim()
        if (!trimmed || /^(?:javascript|data|blob|mailto|tel):/i.test(trimmed)) {
          return null
        }

        return extractRedirectTargetHref(trimmed) ?? normalizeAbsoluteHref(trimmed)
      }

      const isLikelySourceUrl = (value: string | null): value is string => {
        if (!value) {
          return false
        }

        try {
          const url = new URL(value)
          const host = url.hostname.toLowerCase()
          if (!/^https?:$/i.test(url.protocol)) {
            return false
          }
          if (
            host === window.location.hostname.toLowerCase() ||
            host.endsWith('.yiyan.baidu.com') ||
            host === 'eb-static.cdn.bcebos.com' ||
            host === 'ppui-static-wap.cdn.bcebos.com'
          ) {
            return false
          }
          return true
        } catch {
          return false
        }
      }

      const buildSourceLink = (input: unknown) => {
        if (!isObjectRecord(input)) {
          return null
        }

        const url = normalizeHref(
          input.url ??
            input.href ??
            input.link ??
            input.pageUrl ??
            input.page_url ??
            input.webUrl ??
            input.web_url ??
            input.realUrl ??
            input.real_url ??
            input.displayUrl ??
            input.display_url ??
            input.originUrl ??
            input.origin_url ??
            input.targetUrl ??
            input.target_url ??
            input.sourceUrl ??
            input.source_url,
        )
        if (!isLikelySourceUrl(url)) {
          return null
        }

        return {
          url,
          title: normalize(input.title ?? input.name ?? input.text ?? input.label ?? input.desc),
          text: normalize(input.text ?? input.title ?? input.name ?? input.label),
          siteName: normalize(
            input.siteName ?? input.site_name ?? input.site ?? input.source ?? input.domain,
          ),
        }
      }

      const dedupeLinks = (
        items: Array<{
          url: string
          title: string | null
          text: string | null
          siteName: string | null
        }>,
      ) => {
        const seen = new Set<string>()
        const result: typeof items = []
        for (const item of items) {
          if (!item?.url || seen.has(item.url)) {
            continue
          }
          seen.add(item.url)
          result.push(item)
        }
        return result
      }

      const collectSourceLinks = (
        input: unknown,
        result: Array<{
          url: string
          title: string | null
          text: string | null
          siteName: string | null
        }>,
        depth = 0,
      ) => {
        if (depth > 10 || input == null) {
          return
        }

        if (Array.isArray(input)) {
          for (const item of input) {
            collectSourceLinks(item, result, depth + 1)
          }
          return
        }

        const sourceLink = buildSourceLink(input)
        if (sourceLink) {
          result.push(sourceLink)
        }

        if (!isObjectRecord(input)) {
          return
        }

        for (const value of Object.values(input)) {
          if (Array.isArray(value) || isObjectRecord(value)) {
            collectSourceLinks(value, result, depth + 1)
          }
        }
      }

      const collectSourceLinksByFieldPattern = (
        input: unknown,
        pattern: RegExp,
        result: Array<{
          url: string
          title: string | null
          text: string | null
          siteName: string | null
        }>,
        depth = 0,
      ) => {
        if (depth > 10 || input == null) {
          return
        }

        if (Array.isArray(input)) {
          for (const item of input) {
            collectSourceLinksByFieldPattern(item, pattern, result, depth + 1)
          }
          return
        }

        if (!isObjectRecord(input)) {
          return
        }

        for (const [key, value] of Object.entries(input)) {
          if (pattern.test(key)) {
            collectSourceLinks(value, result, depth + 1)
          }

          if (Array.isArray(value) || isObjectRecord(value)) {
            collectSourceLinksByFieldPattern(value, pattern, result, depth + 1)
          }
        }
      }

      const isVisible = (element: Element | null): element is HTMLElement => {
        if (!(element instanceof HTMLElement)) {
          return false
        }

        const style = window.getComputedStyle(element)
        if (
          style.display === 'none' ||
          style.visibility === 'hidden' ||
          style.opacity === '0' ||
          element.hidden
        ) {
          return false
        }

        const rect = element.getBoundingClientRect()
        return rect.width >= 8 && rect.height >= 8
      }

      const extractAnchorLink = (element: Element) => {
        if (!(element instanceof HTMLAnchorElement) || !isVisible(element)) {
          return null
        }

        const url = normalizeHref(element.getAttribute('href') ?? element.href)
        if (!isLikelySourceUrl(url)) {
          return null
        }

        const title = normalize(
          element.getAttribute('title') ??
            element.querySelector('[class*="title"]')?.textContent ??
            element.textContent,
        )
        const siteName = normalize(
          element.getAttribute('data-site-name') ??
            element.dataset.siteName ??
            element.querySelector('[class*="site"]')?.textContent ??
            element.getAttribute('aria-label'),
        )

        return {
          url,
          title: title ?? siteName,
          text: title ?? siteName,
          siteName,
        }
      }

      const urlAttributeNames = [
        'href',
        'data-href',
        'data-url',
        'data-link',
        'data-target-url',
        'data-source-url',
        'data-origin-url',
        'data-real-url',
        'data-page-url',
        'data-web-url',
        'data-redirect',
        'data-jump-url',
        'data-log',
        'data-click',
        'data-extra',
        'data-info',
        'data-value',
      ]

      const firstUrlFromText = (value: string | null): string | null => {
        if (!value) {
          return null
        }

        const httpMatch = value.match(/https?:\/\/[^\s"'<>，。；、]+/i)
        if (httpMatch) {
          return normalizeHref(httpMatch[0])
        }

        const domainMatch = value.match(/\b[a-z0-9-]+(?:\.[a-z0-9-]+)+(?:\/[^\s"'<>，。；、]*)?/i)
        return domainMatch ? normalizeHref(domainMatch[0]) : null
      }

      const readElementUrl = (element: Element): string | null => {
        const candidates: string[] = []
        for (const name of urlAttributeNames) {
          const value = element.getAttribute(name)
          if (value) {
            candidates.push(value)
          }
        }

        if (element instanceof HTMLElement) {
          candidates.push(
            ...Object.values(element.dataset).filter((value): value is string => Boolean(value)),
          )
        }

        for (const candidate of candidates) {
          const url = normalizeHref(candidate) ?? firstUrlFromText(candidate)
          if (isLikelySourceUrl(url)) {
            return url
          }
        }

        const text = normalize(element.textContent ?? '')
        if (text && text.length <= 2_000) {
          const url = firstUrlFromText(text)
          if (isLikelySourceUrl(url)) {
            return url
          }
        }

        return null
      }

      const normalizeShortText = (value: unknown, maxLength = 220): string | null => {
        const text = normalize(value)
        if (!text) {
          return null
        }
        return text.length <= maxLength ? text : null
      }

      const extractElementSourceLink = (element: Element) => {
        if (!isVisible(element)) {
          return null
        }

        const anchorLink = extractAnchorLink(element)
        if (anchorLink) {
          return anchorLink
        }

        const url = readElementUrl(element)
        if (!isLikelySourceUrl(url)) {
          return null
        }

        const card =
          element.closest(
            [
              'a',
              'li',
              'article',
              'section',
              '[role="listitem"]',
              '[class*="item"]',
              '[class*="Item"]',
              '[class*="card"]',
              '[class*="Card"]',
              '[class*="source"]',
              '[class*="Source"]',
              '[class*="reference"]',
              '[class*="Reference"]',
              '[class*="web"]',
              '[class*="Web"]',
            ].join(','),
          ) ?? element

        const title = normalizeShortText(
          card.querySelector('[class*="title"], [class*="Title"], h1, h2, h3, h4')?.textContent ??
            element.getAttribute('title') ??
            element.getAttribute('aria-label') ??
            card.textContent,
        )
        const siteName = normalizeShortText(
          card.querySelector(
            '[class*="site"], [class*="Site"], [class*="domain"], [class*="Domain"], [class*="source"], [class*="Source"]',
          )?.textContent ??
            element.getAttribute('data-site-name') ??
            (element instanceof HTMLElement ? element.dataset.siteName : null),
          80,
        )

        return {
          url,
          title,
          text: title ?? siteName,
          siteName,
        }
      }

      const collectSourceLinksFromElement = (root: HTMLElement) => {
        const candidates = [
          root,
          ...Array.from(
            root.querySelectorAll(
              [
                'a[href]',
                '[data-href]',
                '[data-url]',
                '[data-link]',
                '[data-target-url]',
                '[data-source-url]',
                '[data-origin-url]',
                '[data-real-url]',
                '[data-page-url]',
                '[data-web-url]',
                '[data-log]',
                '[data-click]',
                '[data-extra]',
                '[data-info]',
                'li',
                'article',
                '[role="listitem"]',
                '[class*="item"]',
                '[class*="Item"]',
                '[class*="card"]',
                '[class*="Card"]',
                '[class*="source"]',
                '[class*="Source"]',
                '[class*="reference"]',
                '[class*="Reference"]',
                '[class*="web"]',
                '[class*="Web"]',
              ].join(','),
            ),
          ),
        ]

        return dedupeLinks(
          candidates
            .map((node) => extractElementSourceLink(node))
            .filter((node): node is NonNullable<typeof node> => Boolean(node)),
        )
      }

      const collectVisiblePanelLinks = (triggerText: string | null) => {
        const panelSelectors = [
          '[role="dialog"]',
          '[role="list"]',
          '[class*="modal"]',
          '[class*="Modal"]',
          '[class*="dialog"]',
          '[class*="Dialog"]',
          '[class*="drawer"]',
          '[class*="Drawer"]',
          '[class*="panel"]',
          '[class*="Panel"]',
          '[class*="popover"]',
          '[class*="Popover"]',
          '[class*="popup"]',
          '[class*="Popup"]',
          '[class*="webList"]',
          '[class*="WebList"]',
          '[class*="reference"]',
          '[class*="Reference"]',
          '[class*="material"]',
          '[class*="cite"]',
          '[class*="source"]',
          '[class*="Source"]',
          '[class*="web"]',
          '[class*="Web"]',
          'aside',
          'section',
          'article',
          'body > div',
        ]

        const rootLinks = new Map<HTMLElement, ReturnType<typeof collectSourceLinksFromElement>>()
        const roots = Array.from(document.querySelectorAll(panelSelectors.join(',')))
          .filter((element): element is HTMLElement => isVisible(element))
          .filter((element) => {
            const text = normalize(element.innerText) ?? ''
            const className = String(element.className || '')
            const links = collectSourceLinksFromElement(element)
            rootLinks.set(element, links)
            return (
              links.length > 0 &&
              (/参考|网页|信息源|引用|web pages|references/i.test(text) ||
                Boolean(triggerText && text.includes(triggerText)) ||
                element.getAttribute('role') === 'dialog' ||
                /(drawer|panel|popover|popup|modal|dialog|web|source|reference|cite)/i.test(
                  className,
                ))
            )
          })

        const links = roots.length
          ? roots.flatMap((root) => rootLinks.get(root) ?? collectSourceLinksFromElement(root))
          : []

        return dedupeLinks(links)
      }

      const store = (
        window as typeof window & {
          __NEXT_REDUX_STORE__?: {
            getState?: () => Record<string, unknown>
          }
        }
      ).__NEXT_REDUX_STORE__
      const state = store?.getState?.() ?? {}
      const userInfo = (state.updateReducer as Record<string, unknown> | undefined)?.userInfo
      const selectedModel = (state.model as Record<string, unknown> | undefined)?.selectedModel
      const triggerPattern = new RegExp(triggerPatternSource, 'i')
      const citationFieldPattern = new RegExp(citationFieldPatternSource, 'i')
      const searchFieldPattern = new RegExp(searchFieldPatternSource, 'i')

      const bodyText = normalize(document.body?.innerText) ?? ''
      const loginReason = loginSignals.find((signal) => bodyText.includes(signal)) ?? null
      const reqDone =
        isObjectRecord(userInfo) && typeof userInfo.reqDone === 'boolean' ? userInfo.reqDone : null
      const storeIsLogin =
        isObjectRecord(userInfo) && typeof userInfo.isLogin === 'boolean' ? userInfo.isLogin : null
      const isLogin = storeIsLogin ?? !loginReason

      const modelCandidates: Array<unknown> = []
      if (typeof selectedModel === 'string') {
        modelCandidates.push(selectedModel)
      } else if (isObjectRecord(selectedModel)) {
        modelCandidates.push(
          selectedModel.label,
          selectedModel.name,
          selectedModel.model,
          selectedModel.displayName,
          selectedModel.assistantName,
          selectedModel.assistantKey,
        )
      }
      for (const selector of [
        '[class*="model"][class*="current"]',
        '[class*="currentModel"]',
        '[class*="modelLabel"]',
      ]) {
        modelCandidates.push(document.querySelector(selector)?.textContent ?? null)
      }

      const modelLabel =
        modelCandidates
          .map((candidate) => normalize(candidate))
          .find((candidate): candidate is string => Boolean(candidate)) ?? null

      const editor = document.querySelector(editorSelector)
      const editorText = normalize(
        editor instanceof HTMLElement ? editor.innerText || editor.textContent || '' : '',
      )
      const referenceTriggerText =
        Array.from(document.querySelectorAll('button, [role="button"], a, div, span, p'))
          .filter((element): element is HTMLElement => isVisible(element))
          .map((element) => normalize(element.innerText || element.textContent || ''))
          .find((text): text is string => Boolean(text && triggerPattern.test(text))) ?? null

      const storeReferenceLinks: Array<{
        url: string
        title: string | null
        text: string | null
        siteName: string | null
      }> = []
      collectSourceLinksByFieldPattern(state, citationFieldPattern, storeReferenceLinks)
      collectSourceLinksByFieldPattern(state, searchFieldPattern, storeReferenceLinks)
      const domReferenceLinks = collectVisiblePanelLinks(referenceTriggerText)
      const referenceLinks = dedupeLinks([...storeReferenceLinks, ...domReferenceLinks])

      return {
        url: window.location.href,
        title: normalize(document.title),
        loginRequired: reqDone === true ? !isLogin : Boolean(loginReason && !isLogin),
        loginReason,
        isLogin,
        modelLabel,
        userName: isObjectRecord(userInfo)
          ? normalize(
              (typeof userInfo.uname === 'string' ? userInfo.uname : null) ??
                (typeof userInfo.fuzzyName === 'string' ? userInfo.fuzzyName : null),
            )
          : null,
        editorText,
        bodySample: normalize(bodyText.slice(0, 320)),
        referenceTriggerText,
        referenceLinks,
      }
    },
    {
      editorSelector: WENXIN_EDITOR_SELECTOR,
      loginSignals: WENXIN_LOGIN_SIGNALS,
      triggerPatternSource: WENXIN_REFERENCE_TRIGGER_PATTERN.source,
      citationFieldPatternSource: WENXIN_CITATION_FIELD_PATTERN.source,
      searchFieldPatternSource: WENXIN_SEARCH_FIELD_PATTERN.source,
      redirectQueryKeys: WENXIN_REDIRECT_QUERY_KEYS,
    },
  )
}

async function maybeOpenWenxinReferencePanel(page: PlaywrightPage): Promise<void> {
  const opened = await page
    .evaluate(
      ({ triggerPatternSource }) => {
        const normalize = (value: unknown): string | null => {
          if (typeof value !== 'string') {
            return null
          }
          const trimmed = value
            .replace(/\u00a0/g, ' ')
            .replace(/\r/g, '')
            .replace(/\n{3,}/g, '\n\n')
            .trim()
          return trimmed || null
        }

        const isVisible = (element: Element | null): element is HTMLElement => {
          if (!(element instanceof HTMLElement)) {
            return false
          }
          const style = window.getComputedStyle(element)
          if (
            style.display === 'none' ||
            style.visibility === 'hidden' ||
            style.opacity === '0' ||
            element.hidden
          ) {
            return false
          }
          const rect = element.getBoundingClientRect()
          return rect.width >= 8 && rect.height >= 8
        }

        const triggerPattern = new RegExp(triggerPatternSource, 'i')
        const rawCandidates = Array.from(
          document.querySelectorAll('button, [role="button"], a, [tabindex], div, span, p'),
        )
          .filter((element): element is HTMLElement => isVisible(element))
          .filter((element) => {
            const text = normalize(element.innerText || element.textContent || '')
            return Boolean(text && triggerPattern.test(text))
          })
        const candidates = rawCandidates
          .map((element) => {
            const target = element.closest('button, [role="button"], a, [tabindex]') ?? element
            if (!(target instanceof HTMLElement) || !isVisible(target)) {
              return null
            }

            const rect = target.getBoundingClientRect()
            const text = normalize(target.innerText || target.textContent || '') ?? ''
            const style = window.getComputedStyle(target)
            const className = String(target.className || '')
            let score = 0
            if (target.matches('button, [role="button"], a, [tabindex]')) {
              score += 80
            }
            if (style.cursor === 'pointer' || typeof target.onclick === 'function') {
              score += 40
            }
            if (/(reference|cite|source|web)/i.test(className)) {
              score += 30
            }
            if (text.length <= 80) {
              score += 20
            }
            score -= Math.min((rect.width * rect.height) / 10_000, 50)
            return { target, score }
          })
          .filter((item): item is { target: HTMLElement; score: number } => Boolean(item))

        const seen = new Set<HTMLElement>()
        const trigger = candidates
          .sort((left, right) => right.score - left.score)
          .map((item) => item.target)
          .find((target) => {
            if (seen.has(target)) {
              return false
            }
            seen.add(target)
            return true
          })

        if (!trigger) {
          return false
        }

        trigger.scrollIntoView({ block: 'center', inline: 'center' })
        trigger.click()
        return true
      },
      {
        triggerPatternSource: WENXIN_REFERENCE_TRIGGER_PATTERN.source,
      },
    )
    .catch(() => false)

  if (opened) {
    await page.waitForTimeout(1_000).catch(() => undefined)
    await page.waitForLoadState('networkidle', { timeout: 1_500 }).catch(() => undefined)
  }
}

async function fillWenxinEditorText(page: PlaywrightPage, questionText: string): Promise<void> {
  const editor = page.locator(WENXIN_EDITOR_SELECTOR).first()
  await editor.waitFor({
    state: 'visible',
    timeout: WENXIN_PAGE_READY_TIMEOUT_MS,
  })

  await editor.scrollIntoViewIfNeeded().catch(() => undefined)
  await editor
    .evaluate((node) => {
      if (node instanceof HTMLElement) {
        node.focus()
      }
    })
    .catch(() => undefined)

  let filled = false
  try {
    await editor.fill(questionText)
    filled = true
  } catch {
    filled = false
  }

  const normalizedExpected = questionText.trim().replace(/\s+/g, ' ')
  const normalizedActual = (await editor.innerText().catch(() => '')).trim().replace(/\s+/g, ' ')

  if (!filled || normalizedActual !== normalizedExpected) {
    await page.evaluate(
      ({ selector, text }) => {
        const editorElement = document.querySelector(selector)
        if (!(editorElement instanceof HTMLElement)) {
          throw new Error('wenxin_editor_missing')
        }

        const selection = window.getSelection()
        const range = document.createRange()
        range.selectNodeContents(editorElement)
        selection?.removeAllRanges()
        selection?.addRange(range)

        editorElement.focus()
        try {
          document.execCommand('delete')
          document.execCommand('insertText', false, text)
        } catch {
          // Ignore and fall back to direct DOM mutation below.
        }

        const current = (editorElement.innerText || editorElement.textContent || '')
          .trim()
          .replace(/\s+/g, ' ')
        if (current !== text.trim().replace(/\s+/g, ' ')) {
          editorElement.innerHTML = ''
          editorElement.textContent = text
          editorElement.dispatchEvent(
            new InputEvent('input', {
              bubbles: true,
              inputType: 'insertText',
              data: text,
            }),
          )
          editorElement.dispatchEvent(new Event('change', { bubbles: true }))
        }
      },
      {
        selector: WENXIN_EDITOR_SELECTOR,
        text: questionText,
      },
    )
  }
}

async function submitWenxinQuestion(
  page: PlaywrightPage,
  questionText: string,
  signal: AbortSignal,
): Promise<void> {
  if (signal.aborted) {
    throw new Error('adapter_aborted')
  }

  await fillWenxinEditorText(page, questionText)
  await sleep(150, signal)

  const editor = page.locator(WENXIN_EDITOR_SELECTOR).first()
  await editor.press('Enter').catch(async () => {
    await page.keyboard.press('Enter').catch(() => undefined)
  })
}

async function installWenxinFetchCapture(page: PlaywrightPage): Promise<void> {
  await page.evaluate(
    ({ captureLimit, bodyLimit }) => {
      type CaptureEntry = {
        id: string
        url: string
        method: string
        status: number | null
        contentType: string | null
        requestBody: string | null
        body: string
        done: boolean
        error: string | null
        startedAt: number
        updatedAt: number
      }

      type CaptureState = {
        installed: boolean
        sequence: number
        originalFetch: typeof window.fetch
        entries: CaptureEntry[]
      }

      const globalWindow = window as typeof window & {
        __geoWenxinCapture?: CaptureState
      }

      if (globalWindow.__geoWenxinCapture?.installed) {
        return
      }

      const originalFetch = window.fetch.bind(window)
      const decoder = new TextDecoder()
      const state: CaptureState = {
        installed: true,
        sequence: 0,
        originalFetch,
        entries: [],
      }
      globalWindow.__geoWenxinCapture = state

      window.fetch = (async (...args: [RequestInfo | URL, RequestInit?]) => {
        const request = args[0]
        const init = args[1]
        const rawUrl =
          typeof request === 'string'
            ? request
            : request instanceof URL
              ? request.toString()
              : request && typeof request.url === 'string'
                ? request.url
                : ''
        const url = rawUrl ? new URL(rawUrl, window.location.href).toString() : ''
        const method = (
          init?.method ?? (request instanceof Request ? request.method : 'GET')
        ).toUpperCase()
        const requestBody = typeof init?.body === 'string' ? init.body : null

        const response = await originalFetch(...args)
        if (!url.includes('/eb/chat/conversation/v2')) {
          return response
        }

        const entry: CaptureEntry = {
          id: `wenxin-stream-${(state.sequence += 1)}`,
          url,
          method,
          status: Number.isFinite(response.status) ? response.status : null,
          contentType: response.headers.get('content-type'),
          requestBody,
          body: '',
          done: false,
          error: null,
          startedAt: Date.now(),
          updatedAt: Date.now(),
        }
        state.entries.push(entry)
        if (state.entries.length > captureLimit) {
          state.entries.splice(0, state.entries.length - captureLimit)
        }

        const clone = response.clone()
        void Promise.resolve().then(async () => {
          try {
            if (clone.body) {
              const reader = clone.body.getReader()
              while (true) {
                const { value, done } = await reader.read()
                if (done) {
                  break
                }
                entry.body += decoder.decode(value, { stream: true })
                if (entry.body.length > bodyLimit) {
                  entry.body = entry.body.slice(0, bodyLimit)
                }
                entry.updatedAt = Date.now()
              }
              entry.body += decoder.decode()
            } else {
              entry.body = await clone.text()
            }
          } catch (error) {
            entry.error = error instanceof Error ? `${error.name}: ${error.message}` : String(error)
          } finally {
            if (entry.body.length > bodyLimit) {
              entry.body = entry.body.slice(0, bodyLimit)
            }
            entry.done = true
            entry.updatedAt = Date.now()
          }
        })

        return response
      }) as typeof window.fetch
    },
    {
      captureLimit: WENXIN_STREAM_CAPTURE_LIMIT,
      bodyLimit: WENXIN_STREAM_CAPTURE_BODY_LIMIT,
    },
  )
}

async function resetWenxinFetchCapture(page: PlaywrightPage): Promise<void> {
  await page.evaluate(() => {
    const globalWindow = window as typeof window & {
      __geoWenxinCapture?: {
        entries: WenxinCapturedStreamRecord[]
      }
    }
    if (globalWindow.__geoWenxinCapture) {
      globalWindow.__geoWenxinCapture.entries = []
    }
  })
}

async function readWenxinFetchCapture(page: PlaywrightPage): Promise<WenxinCapturedStreamRecord[]> {
  return await page.evaluate(() => {
    const globalWindow = window as typeof window & {
      __geoWenxinCapture?: {
        entries: WenxinCapturedStreamRecord[]
      }
    }
    return (globalWindow.__geoWenxinCapture?.entries ?? []).map((entry) => ({
      id: entry.id,
      url: entry.url,
      method: entry.method,
      status: entry.status,
      contentType: entry.contentType,
      requestBody: entry.requestBody,
      body: entry.body,
      done: entry.done,
      error: entry.error,
      startedAt: entry.startedAt,
      updatedAt: entry.updatedAt,
    }))
  })
}

function selectWenxinCapture(
  entries: WenxinCapturedStreamRecord[],
  questionText: string,
): WenxinCapturedStreamRecord | null {
  const normalizedQuestion = questionText.trim()
  for (let index = entries.length - 1; index >= 0; index -= 1) {
    const entry = entries[index]
    if (entry.requestBody?.includes(normalizedQuestion)) {
      return entry
    }
  }
  return entries.at(-1) ?? null
}

async function fetchWenxinHistory(
  session: Session,
  sessionId: string,
): Promise<WenxinHistorySummary | null> {
  const payload = await sessionCookieFetchJson<Record<string, unknown>>(
    session,
    WENXIN_HISTORY_URL,
    {
      method: 'POST',
      headers: {
        'content-type': 'application/json',
      },
      body: JSON.stringify({
        sessionId,
        pageSize: 2000,
        timestamp: Date.now(),
        deviceType: 'pc',
      }),
    },
  ).catch(() => null)

  if (!payload) {
    return null
  }

  return parseWenxinHistoryResponse(payload, sessionId)
}

async function waitForWenxinAnswer(
  page: PlaywrightPage,
  session: Session,
  questionText: string,
  signal: AbortSignal,
): Promise<WenxinWaitResult> {
  const startedAt = Date.now()
  let stablePollCount = 0
  let lastSignature: string | null = null
  let latestStream = emptyWenxinStreamSummary()
  let latestHistory: WenxinHistorySummary | null = null
  let latestObservation = mergeWenxinObservation(latestStream, latestHistory)
  let seenCapture = false
  let resolvedSessionId: string | null = null

  while (Date.now() - startedAt <= WENXIN_QUERY_TIMEOUT_MS) {
    if (signal.aborted) {
      throw new Error('adapter_aborted')
    }

    const captures = await readWenxinFetchCapture(page)
    const capture = selectWenxinCapture(captures, questionText)
    if (capture) {
      seenCapture = true
      latestStream = parseWenxinSSEBody(capture.body, capture.done, capture.error)
      resolvedSessionId = latestStream.sessionId ?? resolvedSessionId
    }

    if (!seenCapture && Date.now() - startedAt >= 12_000) {
      latestObservation = mergeWenxinObservation(latestStream, latestHistory)
      return {
        ok: false,
        error: 'wenxin_submit_failed',
        detail: 'no Wenxin conversation stream observed after submission',
        stream: latestStream,
        history: latestHistory,
        observation: latestObservation,
        stablePollCount,
        elapsedMs: Date.now() - startedAt,
      }
    }

    if (resolvedSessionId) {
      latestHistory = await fetchWenxinHistory(session, resolvedSessionId)
    }

    latestObservation = mergeWenxinObservation(latestStream, latestHistory)

    if (latestHistory?.loginRequired || isWenxinLoginMessage(latestObservation.errorMessage)) {
      return {
        ok: false,
        error: 'wenxin_login_required',
        detail: latestObservation.errorMessage ?? latestHistory?.errorMessage ?? 'login_required',
        stream: latestStream,
        history: latestHistory,
        observation: latestObservation,
        stablePollCount,
        elapsedMs: Date.now() - startedAt,
      }
    }

    if (
      isWenxinRiskControlMessage(latestObservation.withdrawText) ||
      isWenxinRiskControlMessage(latestObservation.errorMessage)
    ) {
      return {
        ok: false,
        error: 'wenxin_risk_control',
        detail: latestObservation.withdrawText ?? latestObservation.errorMessage ?? 'risk_control',
        stream: latestStream,
        history: latestHistory,
        observation: latestObservation,
        stablePollCount,
        elapsedMs: Date.now() - startedAt,
      }
    }

    if (
      latestObservation.withdrawText &&
      !latestObservation.answer &&
      !latestObservation.citations.length &&
      !latestObservation.searchResults.length &&
      latestObservation.streamDone
    ) {
      return {
        ok: false,
        error: 'wenxin_withdrawn',
        detail: latestObservation.withdrawText,
        stream: latestStream,
        history: latestHistory,
        observation: latestObservation,
        stablePollCount,
        elapsedMs: Date.now() - startedAt,
      }
    }

    if (
      latestObservation.errorMessage &&
      !latestObservation.answer &&
      !latestObservation.citations.length &&
      !latestObservation.searchResults.length &&
      latestObservation.streamDone
    ) {
      return {
        ok: false,
        error: 'wenxin_stream_error',
        detail: latestObservation.errorMessage,
        stream: latestStream,
        history: latestHistory,
        observation: latestObservation,
        stablePollCount,
        elapsedMs: Date.now() - startedAt,
      }
    }

    if (isObservationComplete(latestObservation)) {
      if (latestObservation.signature === lastSignature) {
        stablePollCount += 1
      } else {
        lastSignature = latestObservation.signature
        stablePollCount = 1
      }

      if (stablePollCount >= WENXIN_STABLE_POLLS_REQUIRED) {
        return {
          ok: true,
          stream: latestStream,
          history: latestHistory,
          observation: latestObservation,
          stablePollCount,
          elapsedMs: Date.now() - startedAt,
        }
      }
    } else {
      stablePollCount = 0
      lastSignature = null
    }

    await sleep(WENXIN_QUERY_POLL_INTERVAL_MS, signal)
  }

  latestObservation = mergeWenxinObservation(latestStream, latestHistory)
  return {
    ok: false,
    error: 'wenxin_query_timeout',
    detail: 'timed out while waiting for Wenxin answer',
    stream: latestStream,
    history: latestHistory,
    observation: latestObservation,
    stablePollCount,
    elapsedMs: Date.now() - startedAt,
  }
}

function buildWenxinCaptureDebug(entries: WenxinCapturedStreamRecord[]): JsonValue[] {
  return entries.slice(-3).map((entry) => ({
    id: entry.id,
    url: entry.url,
    method: entry.method,
    status: entry.status,
    content_type: entry.contentType,
    request_body_preview: entry.requestBody?.slice(0, 240) ?? null,
    body_length: entry.body.length,
    body_preview: entry.body.slice(0, 360) || null,
    done: entry.done,
    error: entry.error ?? null,
  }))
}

export const wenxinAdapter: MonitorAdapter = {
  provider: 'wenxin',
  executionMode: 'playwright',
  async query(context, payload) {
    if (!context.playwright?.page) {
      return {
        status: 'failed',
        summary: '文心一言监测缺少 Playwright 页面上下文。',
        error: buildAdapterError('wenxin_playwright_required', 'wenxin_playwright_required'),
      }
    }

    const questionText = extractQuestionText(payload)
    const page = context.playwright.page

    context.reportProgress('wenxin.page_ready')
    await ensurePageOnWenxinChat(page, context.signal)
    await installWenxinFetchCapture(page)
    await resetWenxinFetchCapture(page)

    const initialSnapshot = await readWenxinPageSnapshot(page)
    if (initialSnapshot.loginRequired) {
      return {
        status: 'failed',
        summary: '文心一言账号未登录或登录态已失效，请先在 desktop client 中重新授权。',
        error: buildAdapterError(
          'wenxin_login_required',
          initialSnapshot.loginReason ?? 'login_required',
          {
            page_url: initialSnapshot.url,
          },
        ),
      }
    }

    context.reportProgress('wenxin.query')
    await submitWenxinQuestion(page, questionText, context.signal)

    context.reportProgress('wenxin.wait_answer')
    const waitResult = await waitForWenxinAnswer(
      page,
      context.session,
      questionText,
      context.signal,
    )
    const captures = await readWenxinFetchCapture(page)
    await maybeOpenWenxinReferencePanel(page).catch(() => undefined)
    const finalSnapshot = await readWenxinPageSnapshot(page).catch(() => initialSnapshot)
    const observation = waitResult.observation
    const answer = normalizeBlockText(observation.answer)
    const pageReferenceLinks = dedupeSourceItems(
      finalSnapshot.referenceLinks
        .map((item) => buildSourceItem(item))
        .filter((item): item is MonitoringSourceItem => Boolean(item)),
    )
    const citations = dedupeSourceItems([...observation.citations, ...pageReferenceLinks])
    const searchResults = dedupeSourceItems([...observation.searchResults, ...pageReferenceLinks])
    const allSources = dedupeSourceItems([...citations, ...searchResults])
    const providerModel = observation.providerModel ?? finalSnapshot.modelLabel ?? 'EB45T'
    const requestID = observation.sessionId

    if (!waitResult.ok) {
      if (waitResult.error === 'wenxin_login_required') {
        return {
          status: 'failed',
          summary: '文心一言账号未登录或登录态已失效，请先在 desktop client 中重新授权。',
          error: buildAdapterError(waitResult.error, waitResult.detail ?? 'login_required', {
            page_url: finalSnapshot.url,
          }),
        }
      }

      if (waitResult.error === 'wenxin_risk_control') {
        return {
          status: 'failed',
          summary: `文心一言请求失败：${waitResult.detail ?? '当前访问环境存在异常，请更换浏览器再尝试提问'}`,
          error: buildAdapterError(waitResult.error, waitResult.detail ?? 'risk_control', {
            page_url: finalSnapshot.url,
          }),
        }
      }

      if (waitResult.error === 'wenxin_query_timeout') {
        return {
          status: 'unknown',
          summary: '文心一言超时未拿到完整答案，已回写 unknown 等待后续补采。',
          error: buildAdapterError(waitResult.error, waitResult.detail ?? waitResult.error, {
            page_url: finalSnapshot.url,
            session_id: requestID,
            stream_done: waitResult.stream.captureDone || waitResult.stream.isEnded,
            history_generating: waitResult.history?.generating ?? null,
          }),
        }
      }

      return {
        status: 'failed',
        summary: `文心一言请求失败：${waitResult.detail ?? waitResult.error}`,
        error: buildAdapterError(waitResult.error, waitResult.detail ?? waitResult.error, {
          page_url: finalSnapshot.url,
          session_id: requestID,
        }),
      }
    }

    if (!answer && !citations.length && !searchResults.length) {
      return {
        status: 'unknown',
        summary: '文心一言返回为空，已回写 unknown 等待后续对账。',
        error: buildAdapterError('wenxin_empty_response', 'wenxin returned no answer or sources'),
      }
    }

    context.reportProgress('wenxin.parse_result')
    return {
      status: 'succeeded',
      summary: '文心一言监控任务执行成功。',
      payload: {
        platform: 'wenxin',
        provider_model: providerModel,
        provider_request_id: observation.providerRequestID,
        request_id: requestID,
        conversation_id: requestID,
        answer,
        citation_count: citations.length,
        search_result_count: searchResults.length,
        source_count: allSources.length,
        citations: toJsonSources(citations),
        references: toJsonSources(allSources),
        search_results: toJsonSources(searchResults),
        sources: toJsonSources(allSources),
        raw_response_json: {
          platform: 'wenxin',
          page_url: finalSnapshot.url,
          page_title: finalSnapshot.title,
          is_login: finalSnapshot.isLogin,
          login_reason: finalSnapshot.loginReason,
          model_label: finalSnapshot.modelLabel,
          user_name: finalSnapshot.userName,
          session_id: requestID,
          answer: answer ?? null,
          answer_block_count: observation.answerBlocks.length,
          answer_blocks: observation.answerBlocks,
          reference_trigger_text: finalSnapshot.referenceTriggerText,
          page_reference_count: pageReferenceLinks.length,
          page_references: toJsonSources(pageReferenceLinks),
          citation_count: citations.length,
          citations: toJsonSources(citations),
          reference_count: allSources.length,
          references: toJsonSources(allSources),
          search_result_count: searchResults.length,
          search_results: toJsonSources(searchResults),
          source_count: allSources.length,
          sources: toJsonSources(allSources),
          stream_summary: toJsonValue(waitResult.stream),
          history_summary: toJsonValue(waitResult.history),
          stable_poll_count: waitResult.stablePollCount,
          elapsed_ms: waitResult.elapsedMs,
          stream_capture_count: captures.length,
          stream_captures: buildWenxinCaptureDebug(captures),
        },
      },
    }
  },
}

export const __wenxinTestUtils = {
  buildSourceItem,
  htmlTableToMarkdown,
  isObservationComplete,
  normalizeUrl,
  parseWenxinHistoryResponse,
  parseWenxinSSEBody,
}