feat: support image prompts across IM channels

src/channels/feishu/bridge.mjs

@@ -1,10 +1,16 @@
 import {
   conversationKey,
+  extractInboundMessage,
   extractText,
   isAllowedSender,
   isBotSender,
   splitText,
 } from './message-utils.mjs';
+import {
+  hasInboundImages,
+  imagePromptUserMessage,
+  promptContentForMessage,
+} from '../shared/image-prompt.mjs';
 import {
   harnessAnswerForQuestion,
   harnessQuestionText,
@@ -21,7 +27,7 @@ const RESOLVED_REPLY_TTL_MS = 30 * 60_000;
 const HELP_TEXT = [
   '北汇星河 AIOS 已连接 DeepSeek Harness。',
   '',
-  '直接发送问题即可继续当前会话。',
+  '直接发送文字或图片即可继续当前会话。',
   '/new  开启一个全新会话',
   '/compact  压缩当前会话的较早上下文',
   '/workspace 工作区绝对路径  切换工作区',
@@ -273,7 +279,8 @@ export class FeishuHarnessBridge {
     await this.#finishReaction(messageId, processingReaction, 'ERROR');
     await this.#send(
       event.message.chat_id,
-      '处理失败，请稍后重试。如果问题持续，请在 DeepSeek Harness 的飞书插件页面检查连接状态。',
+      imagePromptUserMessage(error)
+        ?? '处理失败，请稍后重试。如果问题持续，请在 DeepSeek Harness 的飞书插件页面检查连接状态。',
     ).catch(() => undefined);
   }
 
@@ -297,40 +304,47 @@ export class FeishuHarnessBridge {
       this.#status.messagesReceived += 1;
     }
 
-    const text = extractText(event);
-    if (!text) {
-      await this.#send(event.message.chat_id, '目前仅支持文字消息。');
+    const message = extractInboundMessage(event, this.#client);
+    const text = message.content;
+    const hasImages = hasInboundImages(message);
+    const commandText = event.message.message_type === 'text' && !hasImages ? text : null;
+    if (!text && !hasImages) {
+      await this.#send(event.message.chat_id, '目前支持文字和图片消息。');
       return;
     }
 
-    if (text === '/help') {
+    if (commandText === '/help') {
       await this.#send(event.message.chat_id, HELP_TEXT);
       return;
     }
-    if (text === '/new') {
+    if (commandText === '/new') {
       await this.#state.clearSession(key);
       await this.#send(event.message.chat_id, '已开启全新 Harness 会话。');
       return;
     }
-    if (text === '/status') {
+    if (commandText === '/status') {
       await this.#harness.ensureRunning({ signal: this.#signal });
       await this.#send(event.message.chat_id, '飞书机器人与 DeepSeek Harness 连接正常。');
       return;
     }
-    const workspaceCommand = await runWorkspaceCommand(text, this.#harness, key);
+    const workspaceCommand = commandText === null
+      ? null
+      : await runWorkspaceCommand(text, this.#harness, key);
     if (workspaceCommand) {
       for (const reply of workspaceCommand.messages ?? [workspaceCommand.message]) {
         await this.#send(event.message.chat_id, reply);
       }
       return;
     }
-    const compactCommand = await runCompactCommand(
-      text,
-      this.#harness,
-      this.#state,
-      key,
-      { signal: this.#signal },
-    );
+    const compactCommand = commandText === null
+      ? null
+      : await runCompactCommand(
+          commandText,
+          this.#harness,
+          this.#state,
+          key,
+          { signal: this.#signal },
+        );
     if (compactCommand) {
       await this.#send(event.message.chat_id, compactCommand.message);
       return;
@@ -338,7 +352,7 @@ export class FeishuHarnessBridge {
 
     this.#logger.info?.(`[dsh-feishu] processing ${event.message.chat_type} message ${messageId}`);
     try {
-      await this.#answerWithStream(event, key, text);
+      await this.#answerWithStream(event, key, message);
       this.#status.messagesReplied += 1;
       this.#status.lastReplyAt = new Date().toISOString();
       this.#status.lastError = null;
@@ -362,15 +376,20 @@ export class FeishuHarnessBridge {
     };
   }
 
-  async #answerWithStream(event, key, text) {
+  async #answerWithStream(event, key, message) {
     const chatId = event.message.chat_id;
     const messageId = event.message.message_id;
+    const text = message.content;
+    const content = hasInboundImages(message)
+      ? await promptContentForMessage(message, { signal: this.#signal })
+      : undefined;
     if (!this.#channel?.stream) {
       const { answer } = await askInWorkspaceSession({
         harness: this.#harness,
         state: this.#state,
         key,
         text,
+        content,
         createOptions: { signal: this.#signal },
         existsOptions: { signal: this.#signal },
         askOptions: this.#interactionAskOptions(event, key),
@@ -398,6 +417,7 @@ export class FeishuHarnessBridge {
             state: this.#state,
             key,
             text,
+            content,
             createOptions: { signal: this.#signal },
             existsOptions: { signal: this.#signal },
             askOptions,
@@ -425,6 +445,7 @@ export class FeishuHarnessBridge {
         state: this.#state,
         key,
         text,
+        content,
         createOptions: { signal: this.#signal },
         existsOptions: { signal: this.#signal },
         askOptions: this.#interactionAskOptions(event, key),

src/channels/feishu/message-utils.mjs

@@ -1,3 +1,5 @@
+import { ImagePromptError } from '../shared/image-prompt.mjs';
+
 export function conversationKey(event) {
   const chatType = event?.message?.chat_type;
   if (chatType === 'p2p') {
@@ -10,19 +12,151 @@ export function conversationKey(event) {
   return `group:${chatId}`;
 }
 
-export function extractText(event) {
-  if (event?.message?.message_type !== 'text') return null;
-  let parsed;
+function parsedMessageContent(event) {
+  const value = event?.message?.content;
+  if (value && typeof value === 'object') return value;
+  if (typeof value !== 'string') return null;
   try {
-    parsed = JSON.parse(event.message.content);
+    const parsed = JSON.parse(value);
+    return parsed && typeof parsed === 'object' ? parsed : null;
   } catch {
     return null;
   }
-  let text = typeof parsed.text === 'string' ? parsed.text : '';
-  for (const mention of event.message.mentions ?? []) {
-    if (typeof mention.key === 'string' && mention.key) text = text.replaceAll(mention.key, '');
+}
+
+function withoutMentions(text, event) {
+  let result = typeof text === 'string' ? text : '';
+  for (const mention of event?.message?.mentions ?? []) {
+    if (typeof mention?.key === 'string' && mention.key) {
+      result = result.replaceAll(mention.key, '');
+    }
   }
-  return text.trim();
+  return result.trim();
+}
+
+export function extractText(event) {
+  if (event?.message?.message_type !== 'text') return null;
+  const parsed = parsedMessageContent(event);
+  return parsed ? withoutMentions(parsed.text, event) : null;
+}
+
+function nonEmptyString(value) {
+  return typeof value === 'string' && value.trim() ? value.trim() : null;
+}
+
+function postContent(event, parsed = parsedMessageContent(event)) {
+  if (event?.message?.message_type !== 'post') return null;
+  if (!parsed) return null;
+
+  const lines = [];
+  const title = nonEmptyString(withoutMentions(parsed.title, event));
+  if (title) lines.push(title);
+  const imageKeys = [];
+  for (const paragraph of Array.isArray(parsed.content) ? parsed.content : []) {
+    if (!Array.isArray(paragraph)) continue;
+    let visibleText = '';
+    for (const element of paragraph) {
+      const tag = String(element?.tag ?? '').toLowerCase();
+      if (tag === 'img') {
+        const key = nonEmptyString(element?.image_key);
+        if (key) imageKeys.push(key);
+      } else if (tag === 'text' || tag === 'a' || tag === 'link') {
+        if (typeof element?.text === 'string') visibleText += element.text;
+      }
+    }
+    const line = nonEmptyString(withoutMentions(visibleText, event));
+    if (line) lines.push(line);
+  }
+
+  return {
+    text: lines.join('\n'),
+    imageKeys,
+  };
+}
+
+function headerValue(headers, name) {
+  if (typeof headers?.get === 'function') return headers.get(name);
+  return headers?.[name] ?? headers?.[name.toLowerCase()] ?? null;
+}
+
+function declaredSize(headers) {
+  const header = headerValue(headers, 'content-length');
+  if (header === null || header === undefined || header === '') return null;
+  const value = Number(header);
+  return Number.isFinite(value) && value >= 0 ? value : null;
+}
+
+async function readBoundedStream(stream, { signal, maxBytes }) {
+  if (!stream || typeof stream[Symbol.asyncIterator] !== 'function') {
+    throw new Error('Feishu image download returned no readable stream');
+  }
+  signal?.throwIfAborted();
+  const abort = () => stream.destroy?.(
+    signal.reason ?? new DOMException('Feishu image download aborted', 'AbortError'),
+  );
+  signal?.addEventListener('abort', abort, { once: true });
+  const chunks = [];
+  let size = 0;
+  try {
+    for await (const chunk of stream) {
+      signal?.throwIfAborted();
+      const data = Buffer.from(chunk);
+      size += data.length;
+      if (size > maxBytes) {
+        stream.destroy?.();
+        throw new ImagePromptError(
+          'image-too-large',
+          `Feishu image exceeds ${maxBytes} bytes`,
+          '图片超过 5 MB，请压缩后重试。',
+        );
+      }
+      chunks.push(data);
+    }
+    signal?.throwIfAborted();
+    return Buffer.concat(chunks, size);
+  } finally {
+    signal?.removeEventListener('abort', abort);
+  }
+}
+
+function feishuImageSource(event, client, key) {
+  return {
+    async load({ signal, maxBytes }) {
+      signal?.throwIfAborted();
+      const resource = await client?.im?.v1?.messageResource?.get?.({
+        path: {
+          message_id: event.message.message_id,
+          file_key: key,
+        },
+        params: { type: 'image' },
+      });
+      signal?.throwIfAborted();
+      const size = declaredSize(resource?.headers);
+      if (size !== null && size > maxBytes) {
+        resource?.getReadableStream?.().destroy?.();
+        throw new ImagePromptError(
+          'image-too-large',
+          `Feishu image declares ${size} bytes; the limit is ${maxBytes}`,
+          '图片超过 5 MB，请压缩后重试。',
+        );
+      }
+      return readBoundedStream(resource?.getReadableStream?.(), { signal, maxBytes });
+    },
+  };
+}
+
+export function extractInboundMessage(event, client) {
+  const messageType = event?.message?.message_type;
+  const parsed = parsedMessageContent(event);
+  const post = postContent(event, parsed);
+  const standaloneImageKey = messageType === 'image'
+    ? nonEmptyString(parsed?.image_key)
+    : null;
+  const imageKeys = standaloneImageKey ? [standaloneImageKey] : post?.imageKeys ?? [];
+  return {
+    content: messageType === 'text' ? extractText(event) ?? '' : post?.text ?? '',
+    images: imageKeys.map((key) => feishuImageSource(event, client, key)),
+  };
 }
 
 export function splitText(text, maxChars = 9000) {

src/channels/feishu/plugin-controller.mjs

@@ -5,6 +5,7 @@ export const FEISHU_SECRET_REF = 'DSH_FEISHU_APP_SECRET';
 export const REQUIRED_TENANT_SCOPES = Object.freeze([
   'im:message.p2p_msg:readonly',
   'im:message.group_at_msg:readonly',
+  'im:message:readonly',
   'im:message:send_as_bot',
   'im:message.reactions:write_only',
   'im:message:recall',