@rei-standard/amsg-server 2.5.1 → 2.5.3

package/README.md

@@ -54,15 +54,25 @@ const rei = await createReiServer({
 
 ## 关于 `messageType: 'instant'`
 
-> **Note**：新代码的 instant 消息请用 [@rei-standard/amsg-instant](https://github.com/Tosd0/ReiStandard/blob/main/packages/rei-standard-amsg/instant/README.md)，跳过本端点的"建任务 → 处理 → 删任务" DB 来回。本端点的 `instant` 分支为兼容保留，行为不变、不会有运行时警告。
+> **两条 instant 路径，按各自特点选一条（都是正式支持路径）：**
+> - **本端点的 `messageType: 'instant'`**（create task → process by UUID → delete task）：任务先写进数据库再处理，投递不绑在请求连接上——客户端断开也没关系，任务行还在，能继续跑、能重试，想跑多久跑多久。适合**有数据库、需要长时间生成或保证消息零丢失**的场景。
+> - **[@rei-standard/amsg-instant](https://github.com/Tosd0/ReiStandard/blob/main/packages/rei-standard-amsg/instant/README.md)**：纯 SSE 流 + Web Push backup，不需要数据库，适合无状态边缘运行时（如 Cloudflare Workers）。它的处理挂在响应连接上，客户端一断开就只剩平台给的那点宽限期把活干完（Deno Deploy 实测 ≈20-30s），所以适合**能快速跑完的短即时消息**。
 
 ## AI 接口 `apiUrl` 约束
 
 当 `messageType` 为 `prompted` / `auto`，或 `instant` 使用 AI 配置时：
 
-- `apiUrl` 必须是完整聊天端点（例如：`https://api.openai.com/v1/chat/completions`）。
-- SDK 会自动做最小规范化：去首尾空白、去路径尾部多余 `/`。
-- SDK **不会**自动补全 `/v1`、`/chat/completions` 等路径。
+- `apiUrl` 是聊天端点 URL（例如：`https://api.openai.com/v1/chat/completions`），必须能 `new URL(...)` 解析。
+- SDK 对 OpenAI 风格路径做**幂等**补全（去首尾空白、去尾部多余 `/` 后）：
+
+  | 输入 | 输出 |
+  |---|---|
+  | `https://api.openai.com`（裸域名） | `https://api.openai.com/v1/chat/completions` |
+  | `https://api.openai.com/v1`（版本段结尾） | `https://api.openai.com/v1/chat/completions`（不重复加 `/v1`） |
+  | `https://api.openai.com/v1/chat/completions` | 原样返回 |
+  | `https://api.anthropic.com/v1/messages`（其他自定义路径） | 原样返回，不猜 |
+
+- 规则幂等，传完整 URL 不会被改坏；代理路径很特殊时直接传完整 `…/chat/completions` 绕开补全。
 - `maxTokens` 为可选字段：传了就映射为 `max_tokens`；不传则不指定（由上游模型默认策略决定）。
 
 如果上游返回 `405 Method Not Allowed`，通常表示 `apiUrl` 指向了基础域名而非聊天端点，请优先检查配置值。

package/dist/index.cjs

@@ -27,6 +27,7 @@ async function createAdapter(config) {
 }
 
 // src/server/lib/validation.js
+var _amsgshared = require('@rei-standard/amsg-shared');
 function isValidISO8601(dateString) {
   const date = new Date(dateString);
   return date instanceof Date && !isNaN(date.getTime());
@@ -48,23 +49,6 @@ function isValidUUIDv4(uuid) {
   return uuidV4Regex.test(uuid);
 }
 var VALID_LLM_MESSAGE_ROLES = /* @__PURE__ */ new Set(["system", "user", "assistant", "tool"]);
-var AVATAR_URL_MAX_LENGTH = 2048;
-function validateAvatarUrl(value) {
-  if (value === void 0 || value === null) return null;
-  if (typeof value !== "string") {
-    return "avatarUrl \u5FC5\u987B\u662F\u5B57\u7B26\u4E32";
-  }
-  if (/^data:/i.test(value)) {
-    return "\u5934\u50CF\u4E0D\u652F\u6301\u4F20\u5165 data: URI\uFF0C\u8BF7\u6539\u4E3A\u516C\u7F51\u53EF\u8BBF\u95EE\u7684 https:// \u56FE\u7247 URL";
-  }
-  if (value.length > AVATAR_URL_MAX_LENGTH) {
-    return `\u5934\u50CF URL \u957F\u5EA6 ${value.length} \u5B57\u7B26\u8D85\u8FC7 ${AVATAR_URL_MAX_LENGTH} \u4E0A\u9650\uFF0C\u8BF7\u6539\u4E3A\u66F4\u77ED\u7684\u56FE\u7247 URL`;
-  }
-  if (!isValidUrl(value)) {
-    return "avatarUrl \u4E0D\u662F\u5408\u6CD5 URL";
-  }
-  return null;
-}
 var SPLIT_PATTERN_MAX_LENGTH = 200;
 var SPLIT_PATTERN_MAX_ITEMS = 10;
 function validateSplitPattern(value) {
@@ -188,7 +172,7 @@ function validateScheduleMessagePayload(payload) {
   if (payload.temperature !== void 0 && payload.temperature !== null && (typeof payload.temperature !== "number" || !Number.isFinite(payload.temperature))) {
     return { valid: false, errorCode: "INVALID_PARAMETERS", errorMessage: "\u7F3A\u5C11\u5FC5\u9700\u53C2\u6570\u6216\u53C2\u6570\u683C\u5F0F\u9519\u8BEF", details: { invalidFields: ["temperature (must be a finite number)"] } };
   }
-  const avatarErr = validateAvatarUrl(payload.avatarUrl);
+  const avatarErr = _amsgshared.validateAvatarUrl.call(void 0, payload.avatarUrl);
   if (avatarErr) {
     console.warn("[amsg-server] avatarUrl \u4E0D\u5408\u6CD5\uFF0C\u5DF2\u7F6E\u7A7A\uFF1A", avatarErr);
     payload.avatarUrl = null;
@@ -486,8 +470,11 @@ function isUniqueViolation(error) {
 
 
 
-var _amsgshared = require('@rei-standard/amsg-shared');
+
+
+
 var DEFAULT_SPLIT_REGEX = /([。！？!?]+)/;
+var SLEEP_BETWEEN_MESSAGES_MS = 1500;
 function splitOnceByRegex(chunk, regex) {
   const out = chunk.split(regex).reduce((acc, part, i, arr) => {
     if (i % 2 === 0 && part.trim()) {
@@ -507,38 +494,6 @@ function splitMessageIntoSentences(messageContent, splitPattern = null) {
   }
   return chunks.length > 0 ? chunks : [messageContent];
 }
-function readReasoningContent(llmResponse) {
-  if (!llmResponse || typeof llmResponse !== "object") return null;
-  const choices = (
-    /** @type {{ choices?: unknown }} */
-    llmResponse.choices
-  );
-  if (!Array.isArray(choices) || choices.length === 0) return null;
-  const message = (
-    /** @type {{ message?: { reasoning_content?: unknown, content?: unknown } }} */
-    _optionalChain([choices, 'access', _2 => _2[0], 'optionalAccess', _3 => _3.message])
-  );
-  const raw = _optionalChain([message, 'optionalAccess', _4 => _4.reasoning_content]);
-  if (typeof raw === "string") {
-    const trimmed = raw.trim();
-    if (trimmed.length > 0) return trimmed;
-  }
-  const content = _optionalChain([message, 'optionalAccess', _5 => _5.content]);
-  if (typeof content === "string") {
-    const match = content.match(REASONING_TAG_RE);
-    if (match) {
-      const trimmed = match[2].trim();
-      if (trimmed.length > 0) return trimmed;
-    }
-  }
-  return null;
-}
-var REASONING_TAG_RE = /<(think|thinking|thought)>([\s\S]*?)<\/\1>/i;
-var REASONING_TAG_RE_G = /<(think|thinking|thought)>[\s\S]*?<\/\1>/gi;
-function stripReasoningTags(content) {
-  if (typeof content !== "string" || !content.includes("<")) return content;
-  return content.replace(REASONING_TAG_RE_G, "").trim();
-}
 async function processSingleMessage(task, ctx, providedMasterKey) {
   try {
     const masterKey = providedMasterKey || ctx.masterKey;
@@ -569,9 +524,9 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
     } else {
       throw new Error("Invalid message configuration: no content source available");
     }
-    const reasoning = readReasoningContent(llmResponse);
+    const reasoning = _amsgshared.readReasoningContent.call(void 0, llmResponse);
     if (reasoning) {
-      messageContent = stripReasoningTags(messageContent);
+      messageContent = _amsgshared.stripReasoningTags.call(void 0, messageContent);
     }
     const messages = splitMessageIntoSentences(messageContent, _nullishCoalesce(decryptedPayload.splitPattern, () => ( null)));
     if (!ctx.vapid.email || !ctx.vapid.publicKey || !ctx.vapid.privateKey) {
@@ -599,7 +554,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
         metadata
       });
       await ctx.webpush.sendNotification(pushSubscription, JSON.stringify(reasoningPush));
-      await new Promise((resolve) => setTimeout(resolve, 1500));
+      await new Promise((resolve) => setTimeout(resolve, SLEEP_BETWEEN_MESSAGES_MS));
     }
     for (let i = 0; i < messages.length; i++) {
       const contentPush = _amsgshared.buildContentPush.call(void 0, {
@@ -620,7 +575,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
       });
       await ctx.webpush.sendNotification(pushSubscription, JSON.stringify(contentPush));
       if (i < messages.length - 1) {
-        await new Promise((resolve) => setTimeout(resolve, 1500));
+        await new Promise((resolve) => setTimeout(resolve, SLEEP_BETWEEN_MESSAGES_MS));
       }
     }
     return { success: true, messagesSent: messages.length };
@@ -713,7 +668,7 @@ async function _callAI(payload) {
     );
   }
   const aiData = await aiResponse.json();
-  const content = _optionalChain([aiData, 'optionalAccess', _6 => _6.choices, 'optionalAccess', _7 => _7[0], 'optionalAccess', _8 => _8.message, 'optionalAccess', _9 => _9.content]);
+  const content = _optionalChain([aiData, 'optionalAccess', _2 => _2.choices, 'optionalAccess', _3 => _3[0], 'optionalAccess', _4 => _4.message, 'optionalAccess', _5 => _5.content]);
   if (typeof content !== "string" || !content.trim()) {
     throw new Error("AI API error: response missing choices[0].message.content");
   }
@@ -1170,7 +1125,7 @@ function createUpdateMessageHandler(ctx) {
       }
     }
     if (Object.prototype.hasOwnProperty.call(updates, "avatarUrl")) {
-      const avatarErr = validateAvatarUrl(updates.avatarUrl);
+      const avatarErr = _amsgshared.validateAvatarUrl.call(void 0, updates.avatarUrl);
       if (avatarErr) {
         console.warn("[amsg-server] update-message avatarUrl \u4E0D\u5408\u6CD5\uFF0C\u5DF2\u5FFD\u7565\uFF1A", avatarErr);
         delete updates.avatarUrl;
@@ -1397,7 +1352,7 @@ async function resolveBlobStore(namespace) {
   }
   try {
     const blobModule = await Promise.resolve().then(() => _interopRequireWildcard(require("@netlify/blobs")));
-    const getStore = blobModule.getStore || _optionalChain([blobModule, 'access', _10 => _10.default, 'optionalAccess', _11 => _11.getStore]);
+    const getStore = blobModule.getStore || _optionalChain([blobModule, 'access', _6 => _6.default, 'optionalAccess', _7 => _7.getStore]);
     if (typeof getStore === "function") {
       return getStore({ name: namespace });
     }
@@ -1654,11 +1609,7 @@ function createTenantContextManager(options) {
 }
 
 // src/server/index.js
-function normalizeVapidSubject(email) {
-  const trimmedEmail = String(email || "").trim();
-  if (!trimmedEmail) return "";
-  return /^mailto:/i.test(trimmedEmail) ? trimmedEmail : `mailto:${trimmedEmail}`;
-}
+
 async function createReiServer(config) {
   if (!config) throw new Error("[rei-standard-amsg-server] config is required");
   if (!config.tenant) throw new Error("[rei-standard-amsg-server] tenant config is required");
@@ -1674,7 +1625,7 @@ async function createReiServer(config) {
   const vapid = config.vapid || {};
   if (vapid.email && vapid.publicKey && vapid.privateKey) {
     webpushModule.setVapidDetails(
-      normalizeVapidSubject(vapid.email),
+      _amsgshared.normalizeVapidSubject.call(void 0, vapid.email),
       vapid.publicKey,
       vapid.privateKey
     );
@@ -1731,4 +1682,4 @@ async function createReiServer(config) {
 
 
 
-exports.createAdapter = createAdapter; exports.createReiServer = createReiServer; exports.createTenantToken = createTenantToken; exports.decryptFromStorage = decryptFromStorage; exports.decryptPayload = decryptPayload; exports.deriveUserEncryptionKey = deriveUserEncryptionKey; exports.encryptForStorage = encryptForStorage; exports.isValidISO8601 = isValidISO8601; exports.isValidUUID = isValidUUID; exports.isValidUUIDv4 = isValidUUIDv4; exports.isValidUrl = isValidUrl; exports.validateAvatarUrl = validateAvatarUrl; exports.validateLlmMessagesArray = validateLlmMessagesArray; exports.validateScheduleMessagePayload = validateScheduleMessagePayload; exports.validateSplitPattern = validateSplitPattern; exports.verifyTenantToken = verifyTenantToken;
+exports.createAdapter = createAdapter; exports.createReiServer = createReiServer; exports.createTenantToken = createTenantToken; exports.decryptFromStorage = decryptFromStorage; exports.decryptPayload = decryptPayload; exports.deriveUserEncryptionKey = deriveUserEncryptionKey; exports.encryptForStorage = encryptForStorage; exports.isValidISO8601 = isValidISO8601; exports.isValidUUID = isValidUUID; exports.isValidUUIDv4 = isValidUUIDv4; exports.isValidUrl = isValidUrl; exports.validateAvatarUrl = _amsgshared.validateAvatarUrl; exports.validateLlmMessagesArray = validateLlmMessagesArray; exports.validateScheduleMessagePayload = validateScheduleMessagePayload; exports.validateSplitPattern = validateSplitPattern; exports.verifyTenantToken = verifyTenantToken;

package/dist/index.d.cts

@@ -1,5 +1,6 @@
+import { validateAvatarUrl, readReasoningContent, stripReasoningTags, buildReasoningPush, buildContentPush, normalizeVapidSubject } from '@rei-standard/amsg-shared';
+export { validateAvatarUrl } from '@rei-standard/amsg-shared';
 import { createDecipheriv, createHash, randomBytes, createCipheriv, randomUUID, timingSafeEqual, createHmac } from 'crypto';
-import { buildReasoningPush, buildContentPush } from '@rei-standard/amsg-shared';
 
 /**
  * Adapter Factory
@@ -58,6 +59,7 @@ async function createAdapter(config) {
  * ReiStandard SDK v2.0.1
  */
 
+
 /**
  * Validate ISO 8601 date string.
  * @param {string} dateString
@@ -104,43 +106,12 @@ function isValidUUIDv4(uuid) {
 
 const VALID_LLM_MESSAGE_ROLES = new Set(['system', 'user', 'assistant', 'tool']);
 
-const AVATAR_URL_MAX_LENGTH = 2048;
-
-/**
- * Validate the optional `avatarUrl` field. Rejects `data:` URIs (typically
- * base64-encoded inline images) and anything longer than 2048 chars, both
- * of which are the dominant trigger for downstream 413 / Web Push 4 KB
- * payload errors. Returns an error message string, or null when valid.
- *
- * Mirrors @rei-standard/amsg-instant's `validateAvatarUrl` (kept in lockstep
- * on purpose — both packages forward `avatarUrl` to the same SW push payload).
- *
- * @param {unknown} value
- * @returns {string | null}
- */
-function validateAvatarUrl(value) {
-  if (value === undefined || value === null) return null;
-  if (typeof value !== 'string') {
-    return 'avatarUrl 必须是字符串';
-  }
-  if (/^data:/i.test(value)) {
-    return '头像不支持传入 data: URI，请改为公网可访问的 https:// 图片 URL';
-  }
-  if (value.length > AVATAR_URL_MAX_LENGTH) {
-    return `头像 URL 长度 ${value.length} 字符超过 ${AVATAR_URL_MAX_LENGTH} 上限，请改为更短的图片 URL`;
-  }
-  if (!isValidUrl(value)) {
-    return 'avatarUrl 不是合法 URL';
-  }
-  return null;
-}
-
 const SPLIT_PATTERN_MAX_LENGTH = 200;
 const SPLIT_PATTERN_MAX_ITEMS = 10;
 
 /**
- * Validate the optional `splitPattern` field. Mirrors
- * @rei-standard/amsg-instant's `validateSplitPattern` (kept in lockstep).
+ * Validate the optional `splitPattern` field (amsg-server scheduled tasks
+ * only; amsg-instant 0.8.0 dropped its request-level `splitPattern`).
  * Accepts `string`, `string[]`, or absent/null. Returns an error message
  * string, or null when valid.
  *
@@ -801,13 +772,13 @@ function isUniqueViolation(error) {
  * ReiStandard amsg-server v2.4.0
  *
  * Handles single message content generation and Web Push delivery for
- * scheduled tasks (`fixed` / `prompted` / `auto`) and the legacy
- * via-server instant path (`messageType: 'instant'`).
+ * scheduled tasks (`fixed` / `prompted` / `auto`) and the
+ * in-server instant path (`messageType: 'instant'`).
  *
  * Push wire shape comes from `@rei-standard/amsg-shared`'s
  * discriminated union (`AmsgPush`). The SW (`@rei-standard/amsg-sw`)
  * routes on `messageKind`. Server-driven pushes always carry
- * `source: 'instant'` (for the legacy in-server instant) or
+ * `source: 'instant'` (for the in-server instant path) or
  * `source: 'scheduled'` (for everything else).
  *
  * v2.4.0: when the LLM response carries non-empty
@@ -820,6 +791,11 @@ function isUniqueViolation(error) {
 
 const DEFAULT_SPLIT_REGEX = /([。！？!?]+)/;
 
+// Pacing between consecutive Web Push deliveries (reasoning → content, and
+// between content sentences) so the client renders a natural typing cadence.
+// Kept equal to amsg-instant's SLEEP_BETWEEN_MESSAGES_MS default.
+const SLEEP_BETWEEN_MESSAGES_MS = 1500;
+
 /**
  * Split a single chunk by one regex; on no-match return [chunk] so a later
  * regex in a cascade can still take a swing at it.
@@ -839,11 +815,10 @@ function splitOnceByRegex(chunk, regex) {
 }
 
 /**
- * Sentence splitter — kept byte-for-byte equivalent to
- * `@rei-standard/amsg-instant`'s `splitMessageIntoSentences`. Server carries
- * its own copy to avoid an architectural dependency on the instant package.
- * Name matches the instant export on purpose so cross-package grep finds
- * both copies; if you fix a bug here, fix it in the instant copy too.
+ * Sentence splitter for amsg-server's scheduled `splitPattern` feature
+ * (see standards §6.1). Server-only: amsg-instant 0.8.0 dropped its
+ * request-level `splitPattern`, so there is no instant counterpart to keep
+ * in lockstep.
  *
  * @param {string} messageContent
  * @param {string | string[] | null} [splitPattern=null]
@@ -867,54 +842,6 @@ function splitMessageIntoSentences(messageContent, splitPattern = null) {
   return chunks.length > 0 ? chunks : [messageContent];
 }
 
-/**
- * Read `choices[0].message.reasoning_content` as a non-empty trimmed
- * string, or null when absent / empty. Mirrors
- * `amsg-instant/src/message-processor.js#readReasoningContent`.
- *
- * @param {unknown} llmResponse
- * @returns {string | null}
- */
-function readReasoningContent(llmResponse) {
-  if (!llmResponse || typeof llmResponse !== 'object') return null;
-  const choices = /** @type {{ choices?: unknown }} */ (llmResponse).choices;
-  if (!Array.isArray(choices) || choices.length === 0) return null;
-  const message = /** @type {{ message?: { reasoning_content?: unknown, content?: unknown } }} */ (choices[0])?.message;
-
-  const raw = message?.reasoning_content;
-  if (typeof raw === 'string') {
-    const trimmed = raw.trim();
-    if (trimmed.length > 0) return trimmed;
-  }
-
-  const content = message?.content;
-  if (typeof content === 'string') {
-    const match = content.match(REASONING_TAG_RE);
-    if (match) {
-      const trimmed = match[2].trim();
-      if (trimmed.length > 0) return trimmed;
-    }
-  }
-
-  return null;
-}
-
-const REASONING_TAG_RE = /<(think|thinking|thought)>([\s\S]*?)<\/\1>/i;
-const REASONING_TAG_RE_G = /<(think|thinking|thought)>[\s\S]*?<\/\1>/gi;
-
-/**
- * Mirrors `amsg-instant/src/message-processor.js#stripReasoningTags`.
- * Removes any private chain-of-thought markup leaking through
- * `message.content` so it does not also ship inside ContentPush.
- *
- * @param {string} content
- * @returns {string}
- */
-function stripReasoningTags(content) {
-  if (typeof content !== 'string' || !content.includes('<')) return content;
-  return content.replace(REASONING_TAG_RE_G, '').trim();
-}
-
 /**
  * @typedef {Object} ProcessorContext
  * @property {Object}  webpush           - The web-push module instance (already VAPID-configured).
@@ -1004,7 +931,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
     // `messageId` format — deterministic when we have a task.id so a
     // retry produces the same id for the same (task, sentence) pair
     // (downstream dedupers can key on it). Falls back to a UUID for
-    // the legacy in-server instant path that has no row id.
+    // the in-server instant path that has no row id.
     const messageIdBase = task.id != null
       ? `msg_task_${task.id}`
       : `msg_${randomUUID()}_instant`;
@@ -1028,7 +955,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
         metadata,
       });
       await ctx.webpush.sendNotification(pushSubscription, JSON.stringify(reasoningPush));
-      await new Promise(resolve => setTimeout(resolve, 1500));
+      await new Promise(resolve => setTimeout(resolve, SLEEP_BETWEEN_MESSAGES_MS));
     }
 
     for (let i = 0; i < messages.length; i++) {
@@ -1052,7 +979,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
       await ctx.webpush.sendNotification(pushSubscription, JSON.stringify(contentPush));
 
       if (i < messages.length - 1) {
-        await new Promise(resolve => setTimeout(resolve, 1500));
+        await new Promise(resolve => setTimeout(resolve, SLEEP_BETWEEN_MESSAGES_MS));
       }
     }
 
@@ -1398,12 +1325,20 @@ function createScheduleMessageHandler(ctx) {
     const encryptedPayload = encryptForStorage(JSON.stringify(fullTaskData), userKey);
 
     /**
-     * @deprecated Soft-deprecated. For new code, use @rei-standard/amsg-instant.
-     *   This branch is kept for backward compatibility and the existing behavior
-     *   (create task → process → delete) is unchanged. The dedicated amsg-instant
-     *   package is stateless (no DB roundtrip), deployable to Cloudflare Workers,
-     *   and locks the encryption + push-payload contract behind a single version.
-     *   See packages/rei-standard-amsg/instant/README.md.
+     * In-server instant path. Delivers an instant message through this
+     * server's own task queue (create task → process by UUID → delete task).
+     * The task is written to the database before processing, so delivery is
+     * not tied to the request connection: even if the client disconnects, the
+     * row stays and the generation keeps running (and can be retried) for as
+     * long as it needs. Use this when you have a database and want long or
+     * guaranteed-complete generations with no dropped messages.
+     *
+     * The stateless alternative is `@rei-standard/amsg-instant`: it streams
+     * over SSE with a Web Push backup and needs no database, which makes it a
+     * good fit for edge runtimes (e.g. Cloudflare Workers). Its work rides the
+     * response connection, so after the client disconnects it only has the
+     * platform's brief grace window to finish (≈20-30s observed on Deno
+     * Deploy) — ideal for short instant messages that complete quickly.
      */
     // Instant type: check VAPID before creating the task to avoid orphaned rows
     if (payload.messageType === 'instant') {
@@ -1459,11 +1394,20 @@ function createScheduleMessageHandler(ctx) {
     }
 
     /**
-     * @deprecated Soft-deprecated. For new code, use @rei-standard/amsg-instant.
-     *   The "create-task → process-by-uuid → delete-task" sequence below is
-     *   preserved verbatim so existing clients keep working. New integrations
-     *   should call the dedicated amsg-instant endpoint instead — it skips this
-     *   DB round-trip entirely. See packages/rei-standard-amsg/instant/README.md.
+     * In-server instant path. Delivers an instant message through this
+     * server's own task queue (create task → process by UUID → delete task).
+     * The task is written to the database before processing, so delivery is
+     * not tied to the request connection: even if the client disconnects, the
+     * row stays and the generation keeps running (and can be retried) for as
+     * long as it needs. Use this when you have a database and want long or
+     * guaranteed-complete generations with no dropped messages.
+     *
+     * The stateless alternative is `@rei-standard/amsg-instant`: it streams
+     * over SSE with a Web Push backup and needs no database, which makes it a
+     * good fit for edge runtimes (e.g. Cloudflare Workers). Its work rides the
+     * response connection, so after the client disconnects it only has the
+     * platform's brief grace window to finish (≈20-30s observed on Deno
+     * Deploy) — ideal for short instant messages that complete quickly.
      */
     // Instant type: send immediately
     if (payload.messageType === 'instant') {
@@ -2481,12 +2425,6 @@ function createTenantContextManager(options) {
  */
 
 
-function normalizeVapidSubject(email) {
-  const trimmedEmail = String(email || '').trim();
-  if (!trimmedEmail) return '';
-  return /^mailto:/i.test(trimmedEmail) ? trimmedEmail : `mailto:${trimmedEmail}`;
-}
-
 /**
  * @typedef {Object} VapidConfig
  * @property {string} [email]      - VAPID contact email (e.g. mailto:…).
@@ -2598,4 +2536,4 @@ async function createReiServer(config) {
   };
 }
 
-export { createAdapter, createReiServer, createTenantToken, decryptFromStorage, decryptPayload, deriveUserEncryptionKey, encryptForStorage, isValidISO8601, isValidUUID, isValidUUIDv4, isValidUrl, validateAvatarUrl, validateLlmMessagesArray, validateScheduleMessagePayload, validateSplitPattern, verifyTenantToken };
+export { createAdapter, createReiServer, createTenantToken, decryptFromStorage, decryptPayload, deriveUserEncryptionKey, encryptForStorage, isValidISO8601, isValidUUID, isValidUUIDv4, isValidUrl, validateLlmMessagesArray, validateScheduleMessagePayload, validateSplitPattern, verifyTenantToken };

package/dist/index.d.ts

@@ -1,5 +1,6 @@
+import { validateAvatarUrl, readReasoningContent, stripReasoningTags, buildReasoningPush, buildContentPush, normalizeVapidSubject } from '@rei-standard/amsg-shared';
+export { validateAvatarUrl } from '@rei-standard/amsg-shared';
 import { createDecipheriv, createHash, randomBytes, createCipheriv, randomUUID, timingSafeEqual, createHmac } from 'crypto';
-import { buildReasoningPush, buildContentPush } from '@rei-standard/amsg-shared';
 
 /**
  * Adapter Factory
@@ -58,6 +59,7 @@ async function createAdapter(config) {
  * ReiStandard SDK v2.0.1
  */
 
+
 /**
  * Validate ISO 8601 date string.
  * @param {string} dateString
@@ -104,43 +106,12 @@ function isValidUUIDv4(uuid) {
 
 const VALID_LLM_MESSAGE_ROLES = new Set(['system', 'user', 'assistant', 'tool']);
 
-const AVATAR_URL_MAX_LENGTH = 2048;
-
-/**
- * Validate the optional `avatarUrl` field. Rejects `data:` URIs (typically
- * base64-encoded inline images) and anything longer than 2048 chars, both
- * of which are the dominant trigger for downstream 413 / Web Push 4 KB
- * payload errors. Returns an error message string, or null when valid.
- *
- * Mirrors @rei-standard/amsg-instant's `validateAvatarUrl` (kept in lockstep
- * on purpose — both packages forward `avatarUrl` to the same SW push payload).
- *
- * @param {unknown} value
- * @returns {string | null}
- */
-function validateAvatarUrl(value) {
-  if (value === undefined || value === null) return null;
-  if (typeof value !== 'string') {
-    return 'avatarUrl 必须是字符串';
-  }
-  if (/^data:/i.test(value)) {
-    return '头像不支持传入 data: URI，请改为公网可访问的 https:// 图片 URL';
-  }
-  if (value.length > AVATAR_URL_MAX_LENGTH) {
-    return `头像 URL 长度 ${value.length} 字符超过 ${AVATAR_URL_MAX_LENGTH} 上限，请改为更短的图片 URL`;
-  }
-  if (!isValidUrl(value)) {
-    return 'avatarUrl 不是合法 URL';
-  }
-  return null;
-}
-
 const SPLIT_PATTERN_MAX_LENGTH = 200;
 const SPLIT_PATTERN_MAX_ITEMS = 10;
 
 /**
- * Validate the optional `splitPattern` field. Mirrors
- * @rei-standard/amsg-instant's `validateSplitPattern` (kept in lockstep).
+ * Validate the optional `splitPattern` field (amsg-server scheduled tasks
+ * only; amsg-instant 0.8.0 dropped its request-level `splitPattern`).
  * Accepts `string`, `string[]`, or absent/null. Returns an error message
  * string, or null when valid.
  *
@@ -801,13 +772,13 @@ function isUniqueViolation(error) {
  * ReiStandard amsg-server v2.4.0
  *
  * Handles single message content generation and Web Push delivery for
- * scheduled tasks (`fixed` / `prompted` / `auto`) and the legacy
- * via-server instant path (`messageType: 'instant'`).
+ * scheduled tasks (`fixed` / `prompted` / `auto`) and the
+ * in-server instant path (`messageType: 'instant'`).
  *
  * Push wire shape comes from `@rei-standard/amsg-shared`'s
  * discriminated union (`AmsgPush`). The SW (`@rei-standard/amsg-sw`)
  * routes on `messageKind`. Server-driven pushes always carry
- * `source: 'instant'` (for the legacy in-server instant) or
+ * `source: 'instant'` (for the in-server instant path) or
  * `source: 'scheduled'` (for everything else).
  *
  * v2.4.0: when the LLM response carries non-empty
@@ -820,6 +791,11 @@ function isUniqueViolation(error) {
 
 const DEFAULT_SPLIT_REGEX = /([。！？!?]+)/;
 
+// Pacing between consecutive Web Push deliveries (reasoning → content, and
+// between content sentences) so the client renders a natural typing cadence.
+// Kept equal to amsg-instant's SLEEP_BETWEEN_MESSAGES_MS default.
+const SLEEP_BETWEEN_MESSAGES_MS = 1500;
+
 /**
  * Split a single chunk by one regex; on no-match return [chunk] so a later
  * regex in a cascade can still take a swing at it.
@@ -839,11 +815,10 @@ function splitOnceByRegex(chunk, regex) {
 }
 
 /**
- * Sentence splitter — kept byte-for-byte equivalent to
- * `@rei-standard/amsg-instant`'s `splitMessageIntoSentences`. Server carries
- * its own copy to avoid an architectural dependency on the instant package.
- * Name matches the instant export on purpose so cross-package grep finds
- * both copies; if you fix a bug here, fix it in the instant copy too.
+ * Sentence splitter for amsg-server's scheduled `splitPattern` feature
+ * (see standards §6.1). Server-only: amsg-instant 0.8.0 dropped its
+ * request-level `splitPattern`, so there is no instant counterpart to keep
+ * in lockstep.
  *
  * @param {string} messageContent
  * @param {string | string[] | null} [splitPattern=null]
@@ -867,54 +842,6 @@ function splitMessageIntoSentences(messageContent, splitPattern = null) {
   return chunks.length > 0 ? chunks : [messageContent];
 }
 
-/**
- * Read `choices[0].message.reasoning_content` as a non-empty trimmed
- * string, or null when absent / empty. Mirrors
- * `amsg-instant/src/message-processor.js#readReasoningContent`.
- *
- * @param {unknown} llmResponse
- * @returns {string | null}
- */
-function readReasoningContent(llmResponse) {
-  if (!llmResponse || typeof llmResponse !== 'object') return null;
-  const choices = /** @type {{ choices?: unknown }} */ (llmResponse).choices;
-  if (!Array.isArray(choices) || choices.length === 0) return null;
-  const message = /** @type {{ message?: { reasoning_content?: unknown, content?: unknown } }} */ (choices[0])?.message;
-
-  const raw = message?.reasoning_content;
-  if (typeof raw === 'string') {
-    const trimmed = raw.trim();
-    if (trimmed.length > 0) return trimmed;
-  }
-
-  const content = message?.content;
-  if (typeof content === 'string') {
-    const match = content.match(REASONING_TAG_RE);
-    if (match) {
-      const trimmed = match[2].trim();
-      if (trimmed.length > 0) return trimmed;
-    }
-  }
-
-  return null;
-}
-
-const REASONING_TAG_RE = /<(think|thinking|thought)>([\s\S]*?)<\/\1>/i;
-const REASONING_TAG_RE_G = /<(think|thinking|thought)>[\s\S]*?<\/\1>/gi;
-
-/**
- * Mirrors `amsg-instant/src/message-processor.js#stripReasoningTags`.
- * Removes any private chain-of-thought markup leaking through
- * `message.content` so it does not also ship inside ContentPush.
- *
- * @param {string} content
- * @returns {string}
- */
-function stripReasoningTags(content) {
-  if (typeof content !== 'string' || !content.includes('<')) return content;
-  return content.replace(REASONING_TAG_RE_G, '').trim();
-}
-
 /**
  * @typedef {Object} ProcessorContext
  * @property {Object}  webpush           - The web-push module instance (already VAPID-configured).
@@ -1004,7 +931,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
     // `messageId` format — deterministic when we have a task.id so a
     // retry produces the same id for the same (task, sentence) pair
     // (downstream dedupers can key on it). Falls back to a UUID for
-    // the legacy in-server instant path that has no row id.
+    // the in-server instant path that has no row id.
     const messageIdBase = task.id != null
       ? `msg_task_${task.id}`
       : `msg_${randomUUID()}_instant`;
@@ -1028,7 +955,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
         metadata,
       });
       await ctx.webpush.sendNotification(pushSubscription, JSON.stringify(reasoningPush));
-      await new Promise(resolve => setTimeout(resolve, 1500));
+      await new Promise(resolve => setTimeout(resolve, SLEEP_BETWEEN_MESSAGES_MS));
     }
 
     for (let i = 0; i < messages.length; i++) {
@@ -1052,7 +979,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
       await ctx.webpush.sendNotification(pushSubscription, JSON.stringify(contentPush));
 
       if (i < messages.length - 1) {
-        await new Promise(resolve => setTimeout(resolve, 1500));
+        await new Promise(resolve => setTimeout(resolve, SLEEP_BETWEEN_MESSAGES_MS));
       }
     }
 
@@ -1398,12 +1325,20 @@ function createScheduleMessageHandler(ctx) {
     const encryptedPayload = encryptForStorage(JSON.stringify(fullTaskData), userKey);
 
     /**
-     * @deprecated Soft-deprecated. For new code, use @rei-standard/amsg-instant.
-     *   This branch is kept for backward compatibility and the existing behavior
-     *   (create task → process → delete) is unchanged. The dedicated amsg-instant
-     *   package is stateless (no DB roundtrip), deployable to Cloudflare Workers,
-     *   and locks the encryption + push-payload contract behind a single version.
-     *   See packages/rei-standard-amsg/instant/README.md.
+     * In-server instant path. Delivers an instant message through this
+     * server's own task queue (create task → process by UUID → delete task).
+     * The task is written to the database before processing, so delivery is
+     * not tied to the request connection: even if the client disconnects, the
+     * row stays and the generation keeps running (and can be retried) for as
+     * long as it needs. Use this when you have a database and want long or
+     * guaranteed-complete generations with no dropped messages.
+     *
+     * The stateless alternative is `@rei-standard/amsg-instant`: it streams
+     * over SSE with a Web Push backup and needs no database, which makes it a
+     * good fit for edge runtimes (e.g. Cloudflare Workers). Its work rides the
+     * response connection, so after the client disconnects it only has the
+     * platform's brief grace window to finish (≈20-30s observed on Deno
+     * Deploy) — ideal for short instant messages that complete quickly.
      */
     // Instant type: check VAPID before creating the task to avoid orphaned rows
     if (payload.messageType === 'instant') {
@@ -1459,11 +1394,20 @@ function createScheduleMessageHandler(ctx) {
     }
 
     /**
-     * @deprecated Soft-deprecated. For new code, use @rei-standard/amsg-instant.
-     *   The "create-task → process-by-uuid → delete-task" sequence below is
-     *   preserved verbatim so existing clients keep working. New integrations
-     *   should call the dedicated amsg-instant endpoint instead — it skips this
-     *   DB round-trip entirely. See packages/rei-standard-amsg/instant/README.md.
+     * In-server instant path. Delivers an instant message through this
+     * server's own task queue (create task → process by UUID → delete task).
+     * The task is written to the database before processing, so delivery is
+     * not tied to the request connection: even if the client disconnects, the
+     * row stays and the generation keeps running (and can be retried) for as
+     * long as it needs. Use this when you have a database and want long or
+     * guaranteed-complete generations with no dropped messages.
+     *
+     * The stateless alternative is `@rei-standard/amsg-instant`: it streams
+     * over SSE with a Web Push backup and needs no database, which makes it a
+     * good fit for edge runtimes (e.g. Cloudflare Workers). Its work rides the
+     * response connection, so after the client disconnects it only has the
+     * platform's brief grace window to finish (≈20-30s observed on Deno
+     * Deploy) — ideal for short instant messages that complete quickly.
      */
     // Instant type: send immediately
     if (payload.messageType === 'instant') {
@@ -2481,12 +2425,6 @@ function createTenantContextManager(options) {
  */
 
 
-function normalizeVapidSubject(email) {
-  const trimmedEmail = String(email || '').trim();
-  if (!trimmedEmail) return '';
-  return /^mailto:/i.test(trimmedEmail) ? trimmedEmail : `mailto:${trimmedEmail}`;
-}
-
 /**
  * @typedef {Object} VapidConfig
  * @property {string} [email]      - VAPID contact email (e.g. mailto:…).
@@ -2598,4 +2536,4 @@ async function createReiServer(config) {
   };
 }
 
-export { createAdapter, createReiServer, createTenantToken, decryptFromStorage, decryptPayload, deriveUserEncryptionKey, encryptForStorage, isValidISO8601, isValidUUID, isValidUUIDv4, isValidUrl, validateAvatarUrl, validateLlmMessagesArray, validateScheduleMessagePayload, validateSplitPattern, verifyTenantToken };
+export { createAdapter, createReiServer, createTenantToken, decryptFromStorage, decryptPayload, deriveUserEncryptionKey, encryptForStorage, isValidISO8601, isValidUUID, isValidUUIDv4, isValidUrl, validateLlmMessagesArray, validateScheduleMessagePayload, validateSplitPattern, verifyTenantToken };

package/dist/index.mjs

@@ -27,6 +27,7 @@ async function createAdapter(config) {
 }
 
 // src/server/lib/validation.js
+import { validateAvatarUrl } from "@rei-standard/amsg-shared";
 function isValidISO8601(dateString) {
   const date = new Date(dateString);
   return date instanceof Date && !isNaN(date.getTime());
@@ -48,23 +49,6 @@ function isValidUUIDv4(uuid) {
   return uuidV4Regex.test(uuid);
 }
 var VALID_LLM_MESSAGE_ROLES = /* @__PURE__ */ new Set(["system", "user", "assistant", "tool"]);
-var AVATAR_URL_MAX_LENGTH = 2048;
-function validateAvatarUrl(value) {
-  if (value === void 0 || value === null) return null;
-  if (typeof value !== "string") {
-    return "avatarUrl \u5FC5\u987B\u662F\u5B57\u7B26\u4E32";
-  }
-  if (/^data:/i.test(value)) {
-    return "\u5934\u50CF\u4E0D\u652F\u6301\u4F20\u5165 data: URI\uFF0C\u8BF7\u6539\u4E3A\u516C\u7F51\u53EF\u8BBF\u95EE\u7684 https:// \u56FE\u7247 URL";
-  }
-  if (value.length > AVATAR_URL_MAX_LENGTH) {
-    return `\u5934\u50CF URL \u957F\u5EA6 ${value.length} \u5B57\u7B26\u8D85\u8FC7 ${AVATAR_URL_MAX_LENGTH} \u4E0A\u9650\uFF0C\u8BF7\u6539\u4E3A\u66F4\u77ED\u7684\u56FE\u7247 URL`;
-  }
-  if (!isValidUrl(value)) {
-    return "avatarUrl \u4E0D\u662F\u5408\u6CD5 URL";
-  }
-  return null;
-}
 var SPLIT_PATTERN_MAX_LENGTH = 200;
 var SPLIT_PATTERN_MAX_ITEMS = 10;
 function validateSplitPattern(value) {
@@ -485,9 +469,12 @@ function isUniqueViolation(error) {
 import { randomUUID } from "crypto";
 import {
   buildContentPush,
-  buildReasoningPush
+  buildReasoningPush,
+  readReasoningContent,
+  stripReasoningTags
 } from "@rei-standard/amsg-shared";
 var DEFAULT_SPLIT_REGEX = /([。！？!?]+)/;
+var SLEEP_BETWEEN_MESSAGES_MS = 1500;
 function splitOnceByRegex(chunk, regex) {
   const out = chunk.split(regex).reduce((acc, part, i, arr) => {
     if (i % 2 === 0 && part.trim()) {
@@ -507,38 +494,6 @@ function splitMessageIntoSentences(messageContent, splitPattern = null) {
   }
   return chunks.length > 0 ? chunks : [messageContent];
 }
-function readReasoningContent(llmResponse) {
-  if (!llmResponse || typeof llmResponse !== "object") return null;
-  const choices = (
-    /** @type {{ choices?: unknown }} */
-    llmResponse.choices
-  );
-  if (!Array.isArray(choices) || choices.length === 0) return null;
-  const message = (
-    /** @type {{ message?: { reasoning_content?: unknown, content?: unknown } }} */
-    choices[0]?.message
-  );
-  const raw = message?.reasoning_content;
-  if (typeof raw === "string") {
-    const trimmed = raw.trim();
-    if (trimmed.length > 0) return trimmed;
-  }
-  const content = message?.content;
-  if (typeof content === "string") {
-    const match = content.match(REASONING_TAG_RE);
-    if (match) {
-      const trimmed = match[2].trim();
-      if (trimmed.length > 0) return trimmed;
-    }
-  }
-  return null;
-}
-var REASONING_TAG_RE = /<(think|thinking|thought)>([\s\S]*?)<\/\1>/i;
-var REASONING_TAG_RE_G = /<(think|thinking|thought)>[\s\S]*?<\/\1>/gi;
-function stripReasoningTags(content) {
-  if (typeof content !== "string" || !content.includes("<")) return content;
-  return content.replace(REASONING_TAG_RE_G, "").trim();
-}
 async function processSingleMessage(task, ctx, providedMasterKey) {
   try {
     const masterKey = providedMasterKey || ctx.masterKey;
@@ -599,7 +554,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
         metadata
       });
       await ctx.webpush.sendNotification(pushSubscription, JSON.stringify(reasoningPush));
-      await new Promise((resolve) => setTimeout(resolve, 1500));
+      await new Promise((resolve) => setTimeout(resolve, SLEEP_BETWEEN_MESSAGES_MS));
     }
     for (let i = 0; i < messages.length; i++) {
       const contentPush = buildContentPush({
@@ -620,7 +575,7 @@ async function processSingleMessage(task, ctx, providedMasterKey) {
       });
       await ctx.webpush.sendNotification(pushSubscription, JSON.stringify(contentPush));
       if (i < messages.length - 1) {
-        await new Promise((resolve) => setTimeout(resolve, 1500));
+        await new Promise((resolve) => setTimeout(resolve, SLEEP_BETWEEN_MESSAGES_MS));
       }
     }
     return { success: true, messagesSent: messages.length };
@@ -1654,11 +1609,7 @@ function createTenantContextManager(options) {
 }
 
 // src/server/index.js
-function normalizeVapidSubject(email) {
-  const trimmedEmail = String(email || "").trim();
-  if (!trimmedEmail) return "";
-  return /^mailto:/i.test(trimmedEmail) ? trimmedEmail : `mailto:${trimmedEmail}`;
-}
+import { normalizeVapidSubject } from "@rei-standard/amsg-shared";
 async function createReiServer(config) {
   if (!config) throw new Error("[rei-standard-amsg-server] config is required");
   if (!config.tenant) throw new Error("[rei-standard-amsg-server] tenant config is required");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rei-standard/amsg-server",
-  "version": "2.5.1",
+  "version": "2.5.3",
   "description": "ReiStandard Active Messaging server SDK with pluggable database adapters. Three-axis push schema (messageKind / messageType / messageSubtype) from @rei-standard/amsg-shared. Auto-emits ReasoningPush when the LLM response carries reasoning_content.",
   "repository": {
     "type": "git",
@@ -33,7 +33,7 @@
     "node": ">=20"
   },
   "dependencies": {
-    "@rei-standard/amsg-shared": "0.2.0",
+    "@rei-standard/amsg-shared": "^0.3.0",
     "web-push": "^3.6.7",
     "@netlify/blobs": "^8.1.0"
   },