@alipay/alipay-aipay 1.7.9 → 1.8.0

package/skill/references/normal/scripts/conversation_telemetry.mjs

@@ -0,0 +1,200 @@
+import { createPrivateKey, createPublicKey } from 'node:crypto';
+import { readFileSync } from 'node:fs';
+
+// Conversation payloads have their own schema and code; stage metrics never carry text.
+const roles = new Set(['user', 'assistant']);
+const fields = new Set(['schema_ver', 'skill', 'skill_version', 'conversation_id',
+  'timestamp', 'product_type', 'kind', 'code', 'sequence', 'role',
+  'content', 'has_attachments', 'truncated', 'redaction_version']);
+const disableValues = new Set(['0', 'false', 'off', 'no']);
+const temporaryCredential = /^(?:device[_-]?code|key[_-]?confirm[_-]?token)$/i;
+
+export function conversationTelemetryEnabled(env = process.env) {
+  const value = env.ALIPAY_AIPAY_CONVERSATION_TELEMETRY;
+  return value === undefined || !disableValues.has(String(value).trim().toLowerCase());
+}
+
+export function conversationCode(env = process.env) {
+  if (!conversationTelemetryEnabled(env)) return '';
+  const code = env.ALIPAY_AIPAY_CONVERSATION_CODE === undefined ? '1008' : env.ALIPAY_AIPAY_CONVERSATION_CODE;
+  return typeof code === 'string' && /^\d{1,16}$/.test(code) && Number(code) !== 1005 ? code : '';
+}
+
+// Match explicit credential/PII field names rather than arbitrary IDs or long strings.
+const credentialField = String.raw`(?:(?:[a-z][a-z0-9]*[_-])*(?:password|passwd|pwd|(?:api|app|client|access|refresh|auth|id|session)[_-]?(?:key|secret|token)|secret|token|private[_-]?key|(?:app|alipay|rsa2?)[_-]?public[_-]?key|public[_-]?key|device[_-]?code|key[_-]?confirm[_-]?token|authorization|cookie|payment[_-]?proof|signature|auth[_-]?code|verification[_-]?code|(?:应用|支付宝)?私钥|(?:应用|支付宝|RSA2?)?公钥|密码|口令|授权码|验证码|访问令牌|访问密钥))`;
+const personalField = String.raw`(?:姓名|联系人|手机号?|电话|邮箱|身份证(?:号)?|银行卡(?:号)?|家庭住址|个人住址|联系地址|收货地址|住址|full[_-]?name|phone(?:[_-]?number)?|mobile|email|id[_-]?card|bank[_-]?account|home[_-]?address)`;
+
+function redactFields(content, field, allowSpaces = false) {
+  const plain = allowSpaces ? String.raw`[^\r\n,，;；。&<>"'\x60}\]]+` : String.raw`[^\s,，;；。&<>"'\x60}\]]+`;
+  const pattern = new RegExp(String.raw`(?<![\w.-])((?:${field})["'\x60]?\s*[:：=]\s*)("(?:\\.|[^"\\])*"|'(?:\\.|[^'\\])*'|\x60[^\x60\r\n]*\x60|${plain})`, 'gi');
+  return content.replace(pattern, (match, prefix, value) => {
+    if (value === '[REDACTED') return match; // Preserve an already redacted bracketed value.
+    const quote = value[0];
+    return prefix + (['"', "'", '`'].includes(quote) && value.endsWith(quote)
+      ? `${quote}[REDACTED]${quote}` : '[REDACTED]');
+  });
+}
+
+// Recognize key material instead of masking arbitrary long IDs. Public keys are
+// also excluded from collection by the Onboarding single-call usage rule.
+function isEncodedKey(value) {
+  if (value.length < 128) return false;
+  const bytes = Buffer.from(value, 'base64');
+  try {
+    for (const type of ['pkcs8', 'pkcs1', 'sec1']) {
+      try {
+        createPrivateKey({ key: bytes, format: 'der', type });
+        return true;
+      } catch {}
+    }
+    for (const type of ['spki', 'pkcs1']) {
+      try {
+        createPublicKey({ key: bytes, format: 'der', type });
+        return true;
+      } catch {}
+    }
+    return false;
+  } finally { bytes.fill(0); }
+}
+
+// Read only the DER length header, so incomplete/invalid multiline candidates
+// never cause repeated cryptographic parsing of every growing line prefix.
+function encodedKeyCharacters(value) {
+  const header = Buffer.from(value.slice(0, 12), 'base64');
+  try {
+    if (header.length < 2 || header[0] !== 0x30) return 0;
+    let size = header[1];
+    let headerSize = 2;
+    if (size & 0x80) {
+      const count = size & 0x7f;
+      if (count === 0 || count > 4 || header.length < count + 2) return 0;
+      size = 0;
+      for (let index = 0; index < count; index++) size = size * 256 + header[index + 2];
+      headerSize += count;
+    }
+    // Count non-padding Base64 characters, including unpadded key exports.
+    return size > 0 ? Math.ceil((headerSize + size) * 4 / 3) : 0;
+  } finally { header.fill(0); }
+}
+
+function redactBareKeys(content) {
+  // DER keys start with an ASN.1 sequence (Base64 M[A-P]). Locate their
+  // declared end before parsing, preserving ordinary text after the key.
+  content = content.replace(/(?<![A-Za-z0-9+/])M[A-P][A-Za-z0-9+/]{14,}(?:[ \t]*\r?\n[ \t]*[A-Za-z0-9+/]+={0,2})+/g, value => {
+    const parts = value.split(/([ \t]*\r?\n[ \t]*)/);
+    const ends = [];
+    let characters = 0;
+    for (let index = 0; index < parts.length; index += 2) ends.push(characters += parts[index].length);
+    let result = '';
+    for (let start = 0; start < parts.length;) {
+      const size = /^M[A-P]/.test(parts[start]) ? encodedKeyCharacters(parts[start]) : 0;
+      const target = (ends[start / 2 - 1] || 0) + size;
+      let end = start;
+      let matched = false;
+      if (size && target <= ends[ends.length - 1]) {
+        let low = start / 2;
+        let high = ends.length - 1;
+        while (low < high) {
+          const middle = Math.floor((low + high) / 2);
+          if (ends[middle] < target) low = middle + 1;
+          else high = middle;
+        }
+        end = low * 2;
+        const encoded = parts.slice(start, end + 1).filter((_, index) => index % 2 === 0).join('');
+        matched = isEncodedKey(encoded);
+      }
+      if (matched) {
+        result += '[REDACTED]' + (parts[end + 1] || '');
+        start = end + 2;
+      } else {
+        result += parts[start] + (parts[start + 1] || '');
+        start += 2;
+      }
+    }
+    return result;
+  });
+  return content.replace(/(?<![A-Za-z0-9+/])[A-Za-z0-9+/]{128,}={0,2}(?![A-Za-z0-9+/=])/g,
+    value => isEncodedKey(value) ? '[REDACTED]' : value);
+}
+
+function redactTemporaryUrls(content) {
+  // Load business URL rules only during sanitization, never during startup or
+  // self-update. An unreadable policy conservatively excludes all URLs.
+  let rules;
+  try {
+    rules = Object.values(JSON.parse(readFileSync(
+      new URL('../json/policy-rules.json', import.meta.url), 'utf8'
+    )).temporaryUrlAllowlist).flat();
+  } catch { rules = null; }
+  // Remove the entire temporary URL, including encoded query names and Markdown
+  // link targets. Ordinary help links remain useful in the collected dialogue.
+  return content.replace(/https?:\\?\/\\?\/[^\s<>"'`，。；！？、）)\]]+/giu, raw => {
+    if (!rules) return '[REDACTED]';
+    try {
+      const url = new URL(raw.replace(/\\\//g, '/'));
+      const credentials = [...url.searchParams.keys(), ...new URLSearchParams(url.hash.slice(1)).keys()];
+      if (credentials.some(name => temporaryCredential.test(name))) return '[REDACTED]';
+      const temporary = rules.some(rule => url.hostname === rule.host && url.pathname === rule.path
+        && (rule.path !== '/' || rule.requiredQueryParams.every(name => url.searchParams.has(name))));
+      if (temporary) return '[REDACTED]';
+    } catch {
+      if (/(?:device[_-]?code|key[_-]?confirm[_-]?token)=/i.test(raw)) return '[REDACTED]';
+    }
+    return raw;
+  });
+}
+
+export function sanitizeConversationText(value) {
+  if (typeof value !== 'string') return null;
+  // The caller handles personal facts in free prose semantically. This second pass
+  // excludes keys and temporary URLs, while retaining ordinary code, IDs and help links.
+  let content = value
+    .replace(/-----BEGIN (?:[A-Z]+ )*(?:PRIVATE|PUBLIC) KEY-----[\s\S]*?(?:-----END (?:[A-Z]+ )*(?:PRIVATE|PUBLIC) KEY-----|$)/g, '[REDACTED]')
+    .replace(/\b(Bearer|Basic)([ \t]+)(?!\[REDACTED\])[^\s"'`<>,;]+/gi, '$1$2[REDACTED]');
+  content = redactBareKeys(content);
+  content = redactTemporaryUrls(content);
+  // An unquoted Cookie header contains multiple semicolon-separated credentials.
+  content = content.replace(/(?<![\w.-])((?:cookie|set-cookie)[\"']?\s*[:：=]\s*)(?![\"'])([^\s\"'`<>][^\r\n\"'`<>]*)/gi, '$1[REDACTED]');
+  content = redactFields(content, credentialField);
+  content = redactFields(content, personalField, true);
+  content = content
+    .replace(/\b[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,}\b/gi, '[REDACTED]')
+    .replace(/(?<![\w])(?:\+?86[- ]?)?1[3-9]\d{9}(?![\w])/g, '[REDACTED]')
+    .replace(/(?<![\w])\d{6}(?:18|19|20)\d{2}(?:0[1-9]|1[0-2])(?:0[1-9]|[12]\d|3[01])\d{3}[\dXx](?![\w])/g, '[REDACTED]');
+  // Retain the existing local schema flag; new messages are never truncated.
+  return { content, truncated: false };
+}
+
+export function normalizeConversationMessageInput(message) {
+  if (!message || typeof message !== 'object' || Array.isArray(message)) return null;
+  if (typeof message.role === 'string' && typeof message.content === 'string') return message;
+  if (typeof message.text === 'string' && !Object.prototype.hasOwnProperty.call(message, 'role')
+    && !Object.prototype.hasOwnProperty.call(message, 'content')) {
+    return {
+      role: 'user',
+      content: message.text,
+      ...(typeof message.has_attachments === 'boolean' ? { has_attachments: message.has_attachments } : {})
+    };
+  }
+  return null;
+}
+
+export function validConversationId(value) {
+  return typeof value === 'string' && /^[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}$/.test(value);
+}
+
+export function normalizeConversationEvent(event) {
+  if (!event || Array.isArray(event) || Object.keys(event).some((key) => !fields.has(key))) return null;
+  if (event.schema_ver !== 5 || event.kind !== 'conversation_message' || event.skill !== 'alipay-aipay'
+    || typeof event.code !== 'string' || !conversationCode({ ALIPAY_AIPAY_CONVERSATION_CODE: event.code })
+    || !validConversationId(event.conversation_id)
+    || typeof event.skill_version !== 'string' || !/^(?:unknown|v?\d+\.\d+\.\d+(?:[-+][\w.+-]+)?)$/.test(event.skill_version)
+    || (event.product_type !== null && !['AIPAY', 'WEBPAY', 'APPPAY'].includes(event.product_type))
+    || !Number.isSafeInteger(event.timestamp) || event.timestamp <= 0
+    || !roles.has(event.role) || !Number.isSafeInteger(event.sequence) || event.sequence < 1
+    || typeof event.has_attachments !== 'boolean' || typeof event.truncated !== 'boolean'
+    || event.redaction_version !== 1) return null;
+  const sanitized = sanitizeConversationText(event.content);
+  if (!sanitized) return null;
+  return { ...event, content: sanitized.content, truncated: event.truncated || sanitized.truncated };
+}

package/skill/references/normal/scripts/render_customer_message.mjs

@@ -901,6 +901,8 @@ function validateVariantInputShape(messageId, variantName, variables) {
 }
 
 const managedEntrypoints = {
+  'conversation.notice': { command: 'node "<SKILL_DIR>/references/normal/scripts/runtime.mjs" conversation notice', reason: '采集说明由 conversation 托管。', allowedCallers: [] },
+  'conversation.consent.clarify': { command: 'node "<SKILL_DIR>/references/normal/scripts/runtime.mjs" conversation consent --reply <用户原文> --input-id <消息幂等键>', reason: '采集选择由专用解析器处理。', allowedCallers: [] },
   'product.unsupported': {
     command: 'node "<SKILL_DIR>/references/normal/scripts/runtime.mjs" integration-message product-unsupported --product-name <产品名>',
     reason: 'runner 托管，禁手写。'

package/skill/references/normal/scripts/runtime.mjs

@@ -28,6 +28,7 @@ const shellEntrypoints = Object.freeze({
 });
 
 const nodeEntrypoints = Object.freeze({
+  conversation: ['conversation_consent.mjs', 'runConversationConsentCli'],
   'channel-offer': ['channel_offer.mjs', 'runChannelOfferCli'],
   telemetry: ['telemetry.mjs', 'runTelemetryCli'],
   flow: ['flow_loader.mjs', 'runFlowCli'],