@lightcone-ai/daemon 0.15.72 → 0.15.73

package/mcp-servers/publisher/adapters/kuaishou.js

@@ -56,7 +56,7 @@ export class KuaishouAdapter {
     await this._clickByText('放弃');
     await sleep(500);
     try { await this._cdp.send('Runtime.evaluate', { expression: 'window.scrollTo(0, 300)', returnByValue: false }); } catch {}
-    await this._waitForSelector('input[type="file"], [class*="upload"], [class*="Upload"]', 45000);
+    await this._waitForSelector('input[type="file"], [class*="upload"], [class*="Upload"]', 120000);
 
     const { loggedIn } = await this.checkLoginStatus();
     if (!loggedIn) throw new Error('LOGIN_EXPIRED: 快手登录已过期，请重新扫码连接');
@@ -97,7 +97,7 @@ export class KuaishouAdapter {
 
     // Scroll once to trigger any lazy-rendered upload widgets, then wait
     try { await this._cdp.send('Runtime.evaluate', { expression: 'window.scrollTo(0, 300)', returnByValue: false }); } catch {}
-    await this._waitForSelector('input[type="file"], [class*="upload"], [class*="Upload"]', 45000);
+    await this._waitForSelector('input[type="file"], [class*="upload"], [class*="Upload"]', 120000);
 
     const { loggedIn } = await this.checkLoginStatus();
     if (!loggedIn) throw new Error('LOGIN_EXPIRED: 快手登录已过期，请重新扫码连接');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lightcone-ai/daemon",
-  "version": "0.15.72",
+  "version": "0.15.73",
   "type": "module",
   "main": "src/index.js",
   "bin": {

package/src/chat-bridge.js

@@ -1430,10 +1430,10 @@ server.tool('get_library_file',
 
 // ── record_url_narration ────────────────────────────────────────────────────────
 server.tool('record_url_narration',
-  'Record a silent video of a URL by orchestrating Xvfb + Chromium + ffmpeg, driven by a video plan. Outputs a silent mp4 that can be passed to compose_video_v2 as a video-kind segment with an audio_path for narration.\n\nUse this as the canonical recording step for URL-narration videos. Falls back: if the page needs interactions outside the visual_action vocabulary (clicks, waits, OCR loops), use Monitor (Bash) with custom Playwright instead.\n\nRuntime requirements: this tool only works on a Linux daemon machine with Xvfb + ffmpeg (x11grab) + Chromium installed. macOS / Windows daemons will fail at startup.',
+  'Record a silent video of a URL by driving Chromium on an Xvfb display and capturing it with Playwright recordVideo, driven by a video plan; ffmpeg then transcodes the recording to mp4. Outputs a silent mp4 that can be passed to compose_video_v2 as a video-kind segment with an audio_path for narration.\n\nUse this as the canonical recording step for URL-narration videos. Falls back: if the page needs interactions outside the visual_action vocabulary (clicks, waits, OCR loops), use Monitor (Bash) with custom Playwright instead.\n\nRuntime requirements: this tool only works on a Linux daemon machine with Xvfb + Chromium + ffmpeg installed (ffmpeg is used to transcode the recording to mp4; no x11grab device support needed). macOS / Windows daemons will fail at startup.',
   {
     url: z.string().describe('Page URL to record'),
-    plan: z.record(z.any()).describe('Must be the full output from detail_sections (not plan_video). detail_sections output includes detail_sections_version, sections[], audio metadata, and dwell_ms per phase.'),
+    plan: z.record(z.any()).describe('A video plan: an object with `phases` (or `sections`), each a "visual beat" with `action` (scroll_to_dwell / linear_scroll_during / scroll_back / hold / ...), a target (`target_y` or `focus_region:[y1,y2]`) for scroll-type actions, and `dwell_ms` (how long to hold that beat — should match the segment\'s TTS duration). It can be hand-written or the output of plan_video_segments (whose returned segments array doubles as a valid plan).'),
     output_path: z.string().optional().describe('Workspace-relative output mp4 path. Default tmp/wx3_video/recorded-{ts}.mp4'),
     events_path: z.string().optional().describe('Workspace-relative events.json path. Default ${output_path}.events.json'),
     viewport: z.object({
@@ -1468,7 +1468,7 @@ server.tool('submit_to_library',
     target_platform: z.string().optional().describe('目标发布平台，如 xhs / douyin'),
     metadata: z.record(z.any()).optional().describe('其它 metadata（brand_voice / persona / account / goal_state 等）'),
     understanding: z.record(z.any()).optional().describe('analyze_page 输出'),
-    plan: z.record(z.any()).optional().describe('plan_video / detail_sections 输出'),
+    plan: z.record(z.any()).optional().describe('plan_video_segments 输出（或手写的录屏 plan）'),
   },
   async (args) => {
     if (isBlockedCvmaxEditorVideoTool('submit_to_library')) {
@@ -1529,7 +1529,7 @@ server.tool('request_approval',
     platform:      z.string().describe('Target platform, e.g. "x", "xhs", "email"'),
     description:   z.string().describe('Human-readable summary of what will happen if approved'),
     payload:       z.record(z.any()).describe('Full action parameters (content, media_urls, etc.)'),
-    credential_id: z.string().optional().describe('Which account/credential to use. For publishing, prefer a workspace account_id or real credential UUID. Role aliases like primary/test are accepted only if they uniquely match a workspace account.'),
+    credential_id: z.string().optional().describe('Which account/credential to use. Accepts a workspace account_id, a real credential UUID, the account display name, or a role alias (主号/main/primary, 矩阵号/matrix/secondary, 测试号/test/incubator) — any value works as long as it uniquely matches one workspace account on the target platform. If publishing fails with publish_account_selection_required/ambiguous, pick a value from the returned candidates\' "selectors" list yourself instead of asking the user to re-type an account name.'),
   },
   async ({ action_type, platform, description, payload, credential_id }) => {
     try {

package/src/tools/plan-video-segments.js

@@ -45,6 +45,20 @@ function planDurationSec(audioDurationMs, bufferSec = 0.5) {
   return Math.ceil(raw * 2) / 2; // round up to nearest 0.5s
 }
 
+// Run fn over items with a bounded number of concurrent workers (FIFO drain).
+async function mapWithConcurrency(items, limit, fn) {
+  const queue = items.map((item, index) => ({ item, index }));
+  const workers = Array.from({ length: Math.max(1, Math.min(limit, queue.length)) }, async () => {
+    while (queue.length > 0) {
+      const next = queue.shift();
+      await fn(next.item, next.index);
+    }
+  });
+  await Promise.all(workers);
+}
+
+const TTS_CONCURRENCY = 5;
+
 export async function runPlanVideoSegmentsTool({ segments, workspace_id, voice_id, currentWorkspaceId, api }) {
   if (!Array.isArray(segments) || segments.length === 0) {
     return toolError('segments must be a non-empty array.');
@@ -58,20 +72,31 @@ export async function runPlanVideoSegmentsTool({ segments, workspace_id, voice_i
   const planned = [];
   const errors = [];
 
+  // Synthesize TTS for every text-bearing segment up front, in parallel (bounded),
+  // so an N-segment plan no longer pays N sequential round-trips to the TTS API.
+  const audioResults = new Array(segments.length).fill(null);
+  const ttsJobs = segments
+    .map((seg, i) => ({ i, text: String(seg.text ?? '').trim() }))
+    .filter(job => job.text);
+  await mapWithConcurrency(ttsJobs, TTS_CONCURRENCY, async ({ i, text }) => {
+    try {
+      audioResults[i] = await synthesizeSegmentTts(text, { workspace_id: targetWorkspaceId, voice_id, api });
+    } catch (err) {
+      errors.push(`segments[${i}]: TTS failed — ${err.message}`);
+      audioResults[i] = { audio_path: null, audio_duration_ms: 3000 }; // fallback estimate
+    }
+  });
+  errors.sort((a, b) => {
+    const na = Number((a.match(/segments\[(\d+)\]/) ?? [])[1] ?? 0);
+    const nb = Number((b.match(/segments\[(\d+)\]/) ?? [])[1] ?? 0);
+    return na - nb;
+  });
+
   for (let i = 0; i < segments.length; i++) {
     const seg = segments[i];
     const text = String(seg.text ?? '').trim();
     const kind = String(seg.visual_kind ?? 'image');
-
-    let audioResult = null;
-    if (text) {
-      try {
-        audioResult = await synthesizeSegmentTts(text, { workspace_id: targetWorkspaceId, voice_id, api });
-      } catch (err) {
-        errors.push(`segments[${i}]: TTS failed — ${err.message}`);
-        audioResult = { audio_path: null, audio_duration_ms: 3000 }; // fallback estimate
-      }
-    }
+    const audioResult = audioResults[i];
 
     const audioDurationMs = audioResult?.audio_duration_ms ?? 0;
     let presentation;