@lightcone-ai/daemon 0.15.71 → 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": "0.15.71",
+  "version": "0.15.73",
   "type": "module",
   "main": "src/index.js",
   "bin": {

package/src/_vendor/video/composer-v2/index.js

@@ -9,7 +9,12 @@ const DEFAULT_WIDTH = 1080;
 const DEFAULT_HEIGHT = 1920;
 const DEFAULT_FPS = 30;
 const TRANSITION_DURATION = 0.5;
-const SUBTITLE_FONT = 'PingFang SC,Microsoft YaHei,Arial';
+// ASS `Fontname` is a single family name, not a CSS-style fallback list — a
+// comma here shifts every subsequent field in the `Style:` line, corrupting the
+// whole style so libass renders nothing (i.e. burned-in subtitles look missing).
+// Use one installed family; libass + fontconfig handle glyph fallback. Override
+// via SUBTITLE_FONT env if the deployment ships a different CJK font.
+const SUBTITLE_FONT = (process.env.SUBTITLE_FONT || 'Noto Sans CJK SC').split(',')[0].trim() || 'Noto Sans CJK SC';
 const SUBTITLE_FONT_SIZE = 72;
 const SUBTITLE_MARGIN_V = 120;
 
@@ -23,6 +28,28 @@ function msToAssTimestamp(ms) {
   return `${hr}:${String(min).padStart(2, '0')}:${String(sec).padStart(2, '0')}.${String(cs).padStart(2, '0')}`;
 }
 
+// Split a subtitle block into display-sized sentence units. Breaks on CJK/ASCII
+// sentence punctuation and newlines; merges very short fragments forward so we
+// don't flash one-character lines.
+function splitSubtitleSentences(text) {
+  const raw = String(text ?? '').trim();
+  if (!raw) return [];
+  const pieces = raw
+    .split(/(?<=[。！？!?；;\n])/u)
+    .map(s => s.replace(/\s+/g, ' ').trim())
+    .filter(Boolean);
+  if (pieces.length <= 1) return [raw];
+  const merged = [];
+  for (const piece of pieces) {
+    if (merged.length > 0 && Array.from(merged[merged.length - 1]).length < 6) {
+      merged[merged.length - 1] = `${merged[merged.length - 1]}${piece}`;
+    } else {
+      merged.push(piece);
+    }
+  }
+  return merged;
+}
+
 function wrapSubtitleText(text, maxChars = 14) {
   const chars = Array.from(String(text ?? ''));
   if (chars.length <= maxChars) return chars.join('');
@@ -311,15 +338,29 @@ export async function composeVideoV2({
       }
     }
 
-    // Build subtitle entries with cumulative timeline timestamps
+    // Build subtitle entries with cumulative timeline timestamps. When a clip's
+    // subtitle text spans several sentences, split it into one event per sentence
+    // and spread them across the clip in proportion to their length, so a long
+    // beat reads as sequential lines roughly tracking the narration instead of one
+    // static wall of text.
     let cursorMs = 0;
     const subtitleEntries = [];
     for (const clip of readyClips) {
       if (clip.subtitleText) {
-        subtitleEntries.push({
-          text: clip.subtitleText,
-          start_ms: cursorMs,
-          end_ms: cursorMs + Math.round(clip.duration * 1000),
+        const clipMs = Math.round(clip.duration * 1000);
+        const sentences = splitSubtitleSentences(clip.subtitleText);
+        const totalLen = sentences.reduce((sum, s) => sum + Array.from(s).length, 0) || 1;
+        let offsetMs = 0;
+        sentences.forEach((sentence, idx) => {
+          const share = Array.from(sentence).length / totalLen;
+          const isLast = idx === sentences.length - 1;
+          const spanMs = isLast ? clipMs - offsetMs : Math.max(1, Math.round(clipMs * share));
+          subtitleEntries.push({
+            text: sentence,
+            start_ms: cursorMs + offsetMs,
+            end_ms: cursorMs + offsetMs + spanMs,
+          });
+          offsetMs += spanMs;
         });
       }
       cursorMs += Math.round(clip.duration * 1000);

package/src/_vendor/video/recorder/index.js

@@ -1,17 +1,12 @@
 import { spawn } from 'node:child_process';
 import { mkdirSync } from 'node:fs';
-import { stat, writeFile } from 'node:fs/promises';
+import { mkdtemp, rm, stat, writeFile } from 'node:fs/promises';
+import os from 'node:os';
 import path from 'node:path';
 
 import { launchChromiumMobile, openPageAndSettle } from './chromium-driver.js';
 import { defaultDisplayPool } from './display-pool.js';
-import {
-  createUnexpectedExitWatcher,
-  startFfmpegCapture,
-  stopFfmpegCapture,
-  waitForProcessExit,
-} from './ffmpeg-runner.js';
-import { estimatePlanDurationMs } from './plan-estimator.js';
+import { createUnexpectedExitWatcher, waitForProcessExit } from './ffmpeg-runner.js';
 import { executePlanPhases, normalizePlanPhases } from './plan-executor.js';
 
 const DEFAULT_VIEWPORT = Object.freeze({ width: 1080, height: 1920 });
@@ -177,6 +172,49 @@ function scalePhaseY(phase, zoom) {
   };
 }
 
+// Re-encode the page recording (webm, page content only — no browser chrome) into
+// the mp4 the rest of the pipeline expects, dropping the head segment that covers
+// page load + settle so the clip starts at the first plan phase.
+async function transcodeWebmToMp4({
+  webmPath,
+  outputPath,
+  startMs = 0,
+  fps = DEFAULT_FPS,
+  ffmpegBin = 'ffmpeg',
+} = {}) {
+  const ss = Math.max(0, Number(startMs) || 0) / 1000;
+  const args = [
+    '-y',
+    ...(ss > 0 ? ['-ss', ss.toFixed(3)] : []),
+    '-i', webmPath,
+    '-an',
+    '-c:v', 'libx264',
+    '-preset', 'veryfast',
+    '-pix_fmt', 'yuv420p',
+    ...(Number.isFinite(Number(fps)) && Number(fps) > 0 ? ['-r', String(fps)] : []),
+    '-movflags', '+faststart',
+    outputPath,
+  ];
+  await new Promise((resolve, reject) => {
+    const proc = spawn(ffmpegBin, args, { stdio: ['ignore', 'pipe', 'pipe'] });
+    const errChunks = [];
+    proc.stderr?.on('data', (chunk) => errChunks.push(chunk));
+    proc.once('error', (err) => {
+      const wrapped = new Error(`ffmpeg_spawn_failed:${err.message}`);
+      wrapped.code = 'FFMPEG_SPAWN_FAILED';
+      reject(wrapped);
+    });
+    proc.on('close', (code) => {
+      if (code === 0) return resolve();
+      const wrapped = new Error(
+        `ffmpeg_transcode_failed:code=${code}: ${Buffer.concat(errChunks).toString().slice(-2000)}`
+      );
+      wrapped.code = 'FFMPEG_TRANSCODE_FAILED';
+      reject(wrapped);
+    });
+  });
+}
+
 export async function recordUrlNarration({
   plan,
   output_path,
@@ -189,11 +227,14 @@ export async function recordUrlNarration({
   settle_ms = 4000,
   page_zoom = 1.1,
   displayPool = defaultDisplayPool,
-  ffmpegDurationBufferSec = 8,
   startupProbeMs = 1200,
-  ffmpegStopTimeoutMs = 10000,
   xvfbStopTimeoutMs = 5000,
   postPlanTailMs = 600,
+  recordingDir = null,
+  launchChromiumFn = launchChromiumMobile,
+  openPageFn = openPageAndSettle,
+  transcodeFn = transcodeWebmToMp4,
+  nowMs = () => Date.now(),
 } = {}) {
   const zoom = Number.isFinite(Number(page_zoom)) && Number(page_zoom) > 0 ? Number(page_zoom) : 1.1;
   const rawPhases = normalizePlanPhases(plan);
@@ -212,14 +253,14 @@ export async function recordUrlNarration({
   mkdirSync(path.dirname(resolvedOutputPath), { recursive: true });
   mkdirSync(path.dirname(resolvedEventsPath), { recursive: true });
 
+  const ownTempDir = !recordingDir;
+  const recVideoDir = recordingDir || await mkdtemp(path.join(os.tmpdir(), 'lc-recvid-'));
+
   let displayLease;
   let xvfb;
-  let ffmpeg;
-  let browserSession;
   let xvfbWatcher;
-  let ffmpegWatcher;
+  let browserSession = null;
   let primaryError = null;
-
   const cleanupErrors = [];
 
   try {
@@ -234,11 +275,26 @@ export async function recordUrlNarration({
     });
     xvfbWatcher = createUnexpectedExitWatcher(xvfb.child, 'xvfb');
 
-    browserSession = await launchChromiumMobile({
+    // The page recording captures the page viewport only (no browser chrome),
+    // regardless of the on-screen window. recordVideo starts when the page is
+    // created, so the webm includes goto + settle; we measure that head and trim
+    // it off in transcodeFn.
+    const recordStartedAt = nowMs();
+    browserSession = await launchChromiumFn({
       display,
       viewport: normalizedViewport,
+      contextOptions: {
+        recordVideo: {
+          dir: recVideoDir,
+          size: { width: normalizedViewport.width, height: normalizedViewport.height },
+        },
+      },
     });
-    await openPageAndSettle(browserSession.page, {
+    const videoHandle = typeof browserSession.page.video === 'function'
+      ? browserSession.page.video()
+      : null;
+
+    await openPageFn(browserSession.page, {
       url: resolvedUrl,
       settleMs: settle_ms,
     });
@@ -250,42 +306,53 @@ export async function recordUrlNarration({
       await browserSession.page.waitForTimeout(300);
     }
 
-    const estimatedDurationMs = estimatePlanDurationMs(executablePlan);
-    const estimatedDurationSec = Math.max(
-      5,
-      Math.ceil(estimatedDurationMs / 1000) + Math.max(0, Number(ffmpegDurationBufferSec) || 0)
-    );
-
-    ffmpeg = await startFfmpegCapture({
-      display,
-      outputPath: resolvedOutputPath,
-      width: normalizedViewport.width,
-      height: normalizedViewport.height,
-      fps: normalizedFps,
-      durationSec: estimatedDurationSec,
-      startupProbeMs,
-    });
-    ffmpegWatcher = createUnexpectedExitWatcher(ffmpeg.child, 'ffmpeg');
-
     await scrollToTop(browserSession.page);
     await browserSession.page.waitForTimeout(350);
 
+    const headTrimMs = Math.max(0, nowMs() - recordStartedAt);
+
     const eventsLog = await Promise.race([
       executePlanPhases(browserSession.page, executablePlan),
       xvfbWatcher.promise,
-      ffmpegWatcher.promise,
     ]);
 
     await browserSession.page.waitForTimeout(Math.max(0, Number(postPlanTailMs) || 0));
 
-    ffmpegWatcher.deactivate();
-    await stopFfmpegCapture(ffmpeg, {
-      timeoutMs: ffmpegStopTimeoutMs,
-    });
-
     xvfbWatcher.deactivate();
 
-    await writeFile(resolvedEventsPath, JSON.stringify(eventsLog, null, 2), 'utf8');
+    // Flush the recording: video is written when the context closes.
+    let webmPath = null;
+    try {
+      await browserSession.context.close();
+    } catch (closeError) {
+      cleanupErrors.push(`context_close_failed:${closeError.message}`);
+    }
+    if (videoHandle) {
+      try {
+        webmPath = await videoHandle.path();
+      } catch (pathError) {
+        cleanupErrors.push(`video_path_failed:${pathError.message}`);
+      }
+    }
+    try {
+      await browserSession.browser.close();
+    } catch (closeError) {
+      cleanupErrors.push(`browser_close_failed:${closeError.message}`);
+    }
+    browserSession = null;
+
+    if (!webmPath) {
+      const error = new Error('record_video_not_produced');
+      error.code = 'RECORD_VIDEO_NOT_PRODUCED';
+      throw error;
+    }
+
+    await transcodeFn({
+      webmPath,
+      outputPath: resolvedOutputPath,
+      startMs: headTrimMs,
+      fps: normalizedFps,
+    });
 
     const videoStat = await stat(resolvedOutputPath);
     if (!videoStat.isFile() || videoStat.size <= 0) {
@@ -294,42 +361,36 @@ export async function recordUrlNarration({
       throw error;
     }
 
+    await writeFile(resolvedEventsPath, JSON.stringify(eventsLog, null, 2), 'utf8');
+
+    const lastTms = Array.isArray(eventsLog)
+      ? eventsLog.reduce((max, ev) => Math.max(max, Number(ev?.t_ms) || 0), 0)
+      : 0;
+
     return {
       video_path: resolvedOutputPath,
       events_path: resolvedEventsPath,
       events_log: eventsLog,
+      duration_ms: lastTms > 0 ? lastTms : null,
       display,
     };
   } catch (error) {
     primaryError = error;
     throw error;
   } finally {
-    ffmpegWatcher?.deactivate();
     xvfbWatcher?.deactivate();
 
     if (browserSession) {
       try {
-        await browserSession.close();
+        await browserSession.browser.close();
       } catch (closeError) {
         cleanupErrors.push(`browser_close_failed:${closeError.message}`);
       }
     }
 
-    if (ffmpeg) {
-      try {
-        await stopFfmpegCapture(ffmpeg, {
-          timeoutMs: ffmpegStopTimeoutMs,
-        });
-      } catch (stopError) {
-        cleanupErrors.push(`ffmpeg_stop_failed:${stopError.message}`);
-      }
-    }
-
     if (xvfb) {
       try {
-        await stopXvfb(xvfb, {
-          timeoutMs: xvfbStopTimeoutMs,
-        });
+        await stopXvfb(xvfb, { timeoutMs: xvfbStopTimeoutMs });
       } catch (stopError) {
         cleanupErrors.push(`xvfb_stop_failed:${stopError.message}`);
       }
@@ -339,6 +400,10 @@ export async function recordUrlNarration({
       displayLease.release();
     }
 
+    if (ownTempDir) {
+      await rm(recVideoDir, { recursive: true, force: true }).catch(() => {});
+    }
+
     if (cleanupErrors.length > 0) {
       if (primaryError) {
         primaryError.cleanupErrors = cleanupErrors;

package/src/_vendor/video/recorder/plan-executor.js

@@ -21,6 +21,61 @@ function normalizeRange(value) {
   return [low, high];
 }
 
+// The recorder executes exactly these visual actions. There is no "scroll a bit"
+// blind-scroll action: every scroll phase must say where it lands.
+export const SUPPORTED_PHASE_ACTIONS = Object.freeze([
+  'hold',
+  'smooth_scroll',
+  'fast_scroll',
+  'linear_scroll_during',
+  'scroll_to_dwell',
+  'scroll_back',
+  'cursor_focus',
+]);
+
+// Common spellings authors reach for, mapped onto the canonical action above.
+// Note: scroll_down / scroll_up are intentionally NOT aliased — there is no blind
+// scroll; an unrecognised action raises phase_action_unsupported so the plan gets
+// fixed rather than silently degraded.
+const PHASE_ACTION_ALIASES = new Map([
+  ['scroll_to', 'scroll_to_dwell'],
+  ['scrollto', 'scroll_to_dwell'],
+  ['scroll', 'scroll_to_dwell'],
+  ['scroll_to_region', 'scroll_to_dwell'],
+  ['scroll_to_y', 'scroll_to_dwell'],
+  ['dwell', 'scroll_to_dwell'],
+  ['focus_hold', 'scroll_to_dwell'],
+  ['pan', 'linear_scroll_during'],
+  ['narrated_pan', 'linear_scroll_during'],
+  ['linear_scroll', 'linear_scroll_during'],
+  ['scroll_during', 'linear_scroll_during'],
+  ['scroll_while_narrating', 'linear_scroll_during'],
+  ['return', 'scroll_back'],
+  ['return_anchor', 'scroll_back'],
+  ['back', 'scroll_back'],
+  ['scroll_to_top', 'scroll_back'],
+  ['wait', 'hold'],
+  ['pause', 'hold'],
+  ['stay', 'hold'],
+  ['focus', 'cursor_focus'],
+  ['highlight', 'cursor_focus'],
+]);
+
+function normalizeActionName(rawValue) {
+  const name = normalizeText(rawValue).toLowerCase();
+  if (!name) return '';
+  if (SUPPORTED_PHASE_ACTIONS.includes(name)) return name;
+  return PHASE_ACTION_ALIASES.get(name) || name;
+}
+
+// `visual_action` may be a string (the action name) or an object ({type, target_y, ...}).
+function visualActionObject(section = {}) {
+  const va = section?.visual_action;
+  if (va && typeof va === 'object') return va;
+  if (typeof va === 'string' && va.trim()) return { type: va.trim() };
+  return {};
+}
+
 function inferActionFromCameraMotion(phase = {}) {
   const motion = normalizeText(phase.camera_motion ?? phase.cameraMotion).toLowerCase();
   if (motion === 'narrated_pan') return 'linear_scroll_during';
@@ -30,19 +85,32 @@ function inferActionFromCameraMotion(phase = {}) {
   return '';
 }
 
+function pickFirstNumber(...values) {
+  for (const value of values) {
+    if (value == null) continue;
+    const parsed = Number(value);
+    if (Number.isFinite(parsed)) return Math.round(parsed);
+  }
+  return null;
+}
+
 function normalizeSectionAsPhase(section = {}, index = 0) {
-  const phaseId = normalizeText(section.id ?? section.phase_id) || `phase_${index + 1}`;
-  const visualAction = section.visual_action && typeof section.visual_action === 'object'
-    ? section.visual_action
-    : {};
+  const phaseId = normalizeText(section.id ?? section.phase_id ?? section.name) || `phase_${index + 1}`;
+  const visualAction = visualActionObject(section);
   const focusRegion = normalizeRange(
     section.focus_region
     ?? section.focusRegion
     ?? visualAction.focus_region
     ?? visualAction.focusRegion
   );
-  const explicitAction = normalizeText(section.action ?? visualAction.type).toLowerCase();
-  const action = explicitAction || inferActionFromCameraMotion(section) || 'scroll_to_dwell';
+  const explicitAction = normalizeActionName(section.action ?? visualAction.type ?? visualAction.action);
+  const inferred = explicitAction || inferActionFromCameraMotion(section);
+  const targetY = pickFirstNumber(
+    section.target_y, section.to_y, section.y, section.scroll_y,
+    visualAction.target_y, visualAction.to_y, visualAction.y, visualAction.scroll_y,
+  );
+  const hasTarget = focusRegion != null || targetY != null;
+  const action = inferred || (hasTarget ? 'scroll_to_dwell' : 'hold');
 
   return {
     ...section,
@@ -51,21 +119,42 @@ function normalizeSectionAsPhase(section = {}, index = 0) {
     action,
     focus_region: focusRegion ?? null,
     visual_action: visualAction,
-    target_y: section.target_y ?? visualAction.target_y ?? visualAction.to_y ?? null,
-    from_y: section.from_y ?? visualAction.from_y ?? null,
-    to_y: section.to_y ?? visualAction.to_y ?? null,
-    transition_ms: section.transition_ms ?? visualAction.transition_ms ?? null,
-    duration_ms: section.duration_ms ?? section.dwell_ms ?? null,
+    target_y: targetY,
+    from_y: pickFirstNumber(section.from_y, visualAction.from_y),
+    to_y: pickFirstNumber(section.to_y, visualAction.to_y, section.y, visualAction.y),
+    transition_ms: section.transition_ms ?? visualAction.transition_ms ?? visualAction.duration_ms ?? null,
+    duration_ms: section.duration_ms ?? section.dwell_ms ?? section.audio_duration_ms
+      ?? (section.presentation && Number.isFinite(Number(section.presentation.duration))
+        ? Math.round(Number(section.presentation.duration) * 1000)
+        : null),
   };
 }
 
 export function normalizePlanPhases(plan = {}) {
-  const phases = Array.isArray(plan?.phases) ? plan.phases : [];
-  if (phases.length > 0) return phases;
+  const topLevelPhases = Array.isArray(plan?.phases) ? plan.phases : [];
+  if (topLevelPhases.length > 0) {
+    return topLevelPhases.map((phase, index) => normalizeSectionAsPhase(phase, index));
+  }
 
   const sections = Array.isArray(plan?.sections) ? plan.sections : [];
   if (sections.length > 0) {
-    return sections.map((section, index) => normalizeSectionAsPhase(section, index));
+    const flattened = [];
+    sections.forEach((section, sectionIndex) => {
+      const nested = Array.isArray(section?.phases) ? section.phases : null;
+      if (nested && nested.length > 0) {
+        const prefix = normalizeText(section.id ?? section.phase_id ?? section.name) || `s${sectionIndex + 1}`;
+        nested.forEach((subPhase, subIndex) => {
+          const merged = {
+            ...subPhase,
+            id: subPhase.id ?? subPhase.phase_id ?? subPhase.name ?? `${prefix}_${subIndex + 1}`,
+          };
+          flattened.push(normalizeSectionAsPhase(merged, flattened.length));
+        });
+      } else {
+        flattened.push(normalizeSectionAsPhase(section, flattened.length));
+      }
+    });
+    return flattened;
   }
 
   const error = new Error('plan_phases_required');
@@ -74,13 +163,15 @@ export function normalizePlanPhases(plan = {}) {
 }
 
 function resolvePhaseAction(phase = {}) {
-  const explicit = normalizeText(phase.action ?? phase.visual_action?.type).toLowerCase();
+  const explicit = normalizeActionName(
+    phase.action ?? phase.visual_action?.type ?? phase.visual_action?.action
+  );
   if (explicit) return explicit;
   return inferActionFromCameraMotion(phase);
 }
 
 function resolvePhaseId(phase = {}, index = 0) {
-  return normalizeText(phase.id ?? phase.phase_id) || `phase_${index + 1}`;
+  return normalizeText(phase.id ?? phase.phase_id ?? phase.name) || `phase_${index + 1}`;
 }
 
 function nowMs(getNowMs) {
@@ -94,9 +185,12 @@ function resolveTransitionMs(phase, fallback) {
 }
 
 function resolveTargetY(phase, fallback = null) {
-  const raw = phase?.target_y ?? phase?.to_y ?? phase?.visual_action?.target_y ?? phase?.visual_action?.to_y;
-  const parsed = Number(raw);
-  if (Number.isFinite(parsed)) return Math.round(parsed);
+  const explicit = pickFirstNumber(
+    phase?.target_y, phase?.to_y, phase?.y, phase?.scroll_y,
+    phase?.visual_action?.target_y, phase?.visual_action?.to_y,
+    phase?.visual_action?.y, phase?.visual_action?.scroll_y,
+  );
+  if (explicit != null) return explicit;
 
   const focusRegion = normalizeRange(
     phase?.focus_region
@@ -113,6 +207,20 @@ function resolveTargetY(phase, fallback = null) {
   return fallback;
 }
 
+function requireTargetY(phase, action) {
+  const targetY = resolveTargetY(phase, null);
+  if (targetY == null) {
+    const error = new Error(
+      `phase_target_y_required: phase "${resolvePhaseId(phase)}" uses "${action}" but has no `
+      + 'target_y / to_y / y or focus_region — every scroll phase must say where it lands '
+      + '(there is no blind scroll)',
+    );
+    error.code = 'PHASE_TARGET_Y_REQUIRED';
+    throw error;
+  }
+  return targetY;
+}
+
 function resolveFromY(phase, fallback = null) {
   const raw = phase?.from_y ?? phase?.visual_action?.from_y;
   const parsed = Number(raw);
@@ -192,8 +300,8 @@ async function executeHold(page, phase) {
   return { anchorY: null };
 }
 
-async function executeSmoothScroll(page, phase, { fallbackTargetY = null } = {}) {
-  const targetY = resolveTargetY(phase, fallbackTargetY);
+async function executeSmoothScroll(page, phase) {
+  const targetY = requireTargetY(phase, 'smooth_scroll');
   const transitionMs = resolveTransitionMs(phase, 900);
   await animateScroll(page, {
     targetY,
@@ -205,8 +313,8 @@ async function executeSmoothScroll(page, phase, { fallbackTargetY = null } = {})
   return { anchorY: targetY };
 }
 
-async function executeFastScroll(page, phase, { fallbackTargetY = null } = {}) {
-  const targetY = resolveTargetY(phase, fallbackTargetY);
+async function executeFastScroll(page, phase) {
+  const targetY = requireTargetY(phase, 'fast_scroll');
   const transitionMs = resolveTransitionMs(phase, 420);
   await animateScroll(page, {
     targetY,
@@ -218,12 +326,9 @@ async function executeFastScroll(page, phase, { fallbackTargetY = null } = {}) {
   return { anchorY: targetY };
 }
 
-async function executeLinearScrollDuring(page, phase, {
-  fallbackFromY = null,
-  fallbackTargetY = null,
-} = {}) {
+async function executeLinearScrollDuring(page, phase, { fallbackFromY = null } = {}) {
   const fromY = resolveFromY(phase, fallbackFromY);
-  const toY = resolveTargetY(phase, fallbackTargetY);
+  const toY = requireTargetY(phase, 'linear_scroll_during');
   const durationMs = resolveDurationMs(phase, null);
   if (!Number.isFinite(Number(durationMs)) || Number(durationMs) <= 0) {
     const error = new Error('linear_scroll_duration_required');
@@ -247,8 +352,8 @@ async function executeLinearScrollDuring(page, phase, {
   return { anchorY: toY };
 }
 
-async function executeScrollToDwell(page, phase, { fallbackTargetY = null } = {}) {
-  const targetY = resolveTargetY(phase, fallbackTargetY);
+async function executeScrollToDwell(page, phase) {
+  const targetY = requireTargetY(phase, 'scroll_to_dwell');
   const transitionMs = resolveTransitionMs(phase, 820);
   await animateScroll(page, {
     targetY,
@@ -286,8 +391,8 @@ async function executeScrollBack(page, phase, { fallbackTargetY = 0 } = {}) {
   return { anchorY: targetY };
 }
 
-async function executeCursorFocus(page, phase, { fallbackTargetY = null } = {}) {
-  const targetY = resolveTargetY(phase, fallbackTargetY);
+async function executeCursorFocus(page, phase) {
+  const targetY = requireTargetY(phase, 'cursor_focus');
   const transitionMs = resolveTransitionMs(phase, 650);
   await animateScroll(page, {
     targetY,
@@ -313,34 +418,34 @@ async function executePhase(page, phase, {
   initialAnchorY = 0,
 } = {}) {
   const action = resolvePhaseAction(phase);
-  const fallbackY = lastAnchorY ?? initialAnchorY;
+  const fallbackFromY = lastAnchorY ?? initialAnchorY;
 
   if (action === 'hold') {
     return executeHold(page, phase);
   }
   if (action === 'smooth_scroll') {
-    return executeSmoothScroll(page, phase, { fallbackTargetY: fallbackY });
+    return executeSmoothScroll(page, phase);
   }
   if (action === 'fast_scroll') {
-    return executeFastScroll(page, phase, { fallbackTargetY: fallbackY });
+    return executeFastScroll(page, phase);
   }
   if (action === 'linear_scroll_during') {
-    return executeLinearScrollDuring(page, phase, {
-      fallbackFromY: fallbackY,
-      fallbackTargetY: fallbackY,
-    });
+    return executeLinearScrollDuring(page, phase, { fallbackFromY });
   }
   if (action === 'scroll_to_dwell') {
-    return executeScrollToDwell(page, phase, { fallbackTargetY: fallbackY });
+    return executeScrollToDwell(page, phase);
   }
   if (action === 'scroll_back') {
     return executeScrollBack(page, phase, { fallbackTargetY: 0 });
   }
   if (action === 'cursor_focus') {
-    return executeCursorFocus(page, phase, { fallbackTargetY: fallbackY });
+    return executeCursorFocus(page, phase);
   }
 
-  const error = new Error(`phase_action_unsupported:${action || 'empty'}`);
+  const error = new Error(
+    `phase_action_unsupported:${action || 'empty'} — supported actions: ${SUPPORTED_PHASE_ACTIONS.join(', ')}`
+    + ' (there is no blind scroll_down/scroll_up; use scroll_to_dwell with target_y or focus_region)',
+  );
   error.code = 'PHASE_ACTION_UNSUPPORTED';
   throw error;
 }

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;
@@ -87,11 +112,19 @@ export async function runPlanVideoSegmentsTool({ segments, workspace_id, voice_i
       presentation = { duration, ...(kind === 'scroll' ? { style: 'scroll' } : {}) };
     }
 
+    // dwell_ms lets the same segment double as a record_url_narration plan phase
+    // (the recorder reads dwell_ms / duration_ms for how long to hold each beat).
+    // Prefer the real measured audio length; fall back to the planned visual duration.
+    const dwellMs = audioDurationMs > 0
+      ? audioDurationMs
+      : Math.round((presentation.duration ?? presentation.per_card_duration ?? 4) * 1000);
+
     const planned_seg = {
       ...seg,
       ...(audioResult?.audio_path ? { audio_path: audioResult.audio_path } : {}),
       ...(text ? { subtitle_text: text } : {}),
       presentation: { ...presentation, ...(seg.presentation ?? {}) },
+      dwell_ms: seg.dwell_ms ?? dwellMs,
     };
     if (audioResult?.audio_duration_ms) {
       planned_seg.audio_duration_ms = audioResult.audio_duration_ms;