@lightcone-ai/daemon 0.15.28 → 0.15.30

package/mcp-servers/official/video-narration-planner/core.js

@@ -1443,6 +1443,7 @@ export async function detailSections({
   const totalDurationMs = sections.reduce((sum, item) => sum + item.dwell_ms, 0);
 
   return {
+    detail_sections_version: 1,
     sections,
     outro_video_id: toSafeString(strategy?.outro_video_id || strategy?.outroVideoId) || DEFAULT_OUTRO_VIDEO_ID,
     total_duration_ms: totalDurationMs,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lightcone-ai/daemon",
-  "version": "0.15.28",
+  "version": "0.15.30",
   "type": "module",
   "main": "src/index.js",
   "bin": {

package/src/_vendor/video/recorder/chromium-driver.js

@@ -38,7 +38,14 @@ export async function launchChromiumMobile({
   hasTouch = true,
   headless = false,
   channel = 'chrome',
-  launchArgs = ['--no-sandbox', '--disable-dev-shm-usage'],
+  launchArgs = [
+    '--no-sandbox',
+    '--disable-dev-shm-usage',
+    '--kiosk',
+    '--disable-infobars',
+    '--no-first-run',
+    '--no-default-browser-check',
+  ],
   playwrightModule = 'playwright',
   launchOptions = {},
   contextOptions = {},

package/src/chat-bridge.js

@@ -1472,7 +1472,7 @@ 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 plus an events.json timestamp log that compose_video can use to align audio segments.\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.',
   {
     url: z.string().describe('Page URL to record'),
-    plan: z.record(z.any()).describe('Video plan from plan_video / detail_sections — containing phases array with visual_action per phase'),
+    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.'),
     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({
@@ -1481,7 +1481,6 @@ server.tool('record_url_narration',
     }).optional().describe('Default 1080x1920 (mobile portrait). Override only if the plan requires a different shape.'),
     fps: z.number().optional().describe('Default 30. Do not change unless needed.'),
     settle_ms: z.number().optional().describe('Default 4000. Settle wait after navigation before recording starts.'),
-    page_zoom: z.number().optional().describe('Browser zoom factor applied before recording. Default 1.1 (10% zoom in). Set to 1.0 to disable. Plan Y coordinates are automatically scaled by this factor.'),
   },
   async (args) => {
     if (isBlockedCvmaxEditorVideoTool('record_url_narration')) {
@@ -1512,7 +1511,7 @@ server.tool('compose_video',
       text: z.string().describe('Subtitle text for this segment (the narration sentence).'),
       start_ms: z.number().describe('Subtitle start time in milliseconds.'),
       end_ms: z.number().describe('Subtitle end time in milliseconds.'),
-    })).optional().describe('Subtitle segments to burn into the video. Pass each phase narration text with its start/end time (derived from detail_sections dwell_ms). Omit to produce no subtitles.'),
+    })).optional().describe('Subtitle segments to burn into the video. Pass each phase sentence text (from detail_sections sections[].sentence) with cumulative start/end time derived from dwell_ms. Omit to produce no subtitles.'),
     outro_path: z.string().optional().describe('Optional outro mp4 path. If omitted, uses ~/.lightcone/assets/outros/default.mp4 when present.'),
     target: z.enum(['short_video_cn', 'douyin', 'xhs']).optional().describe('Transcode target profile. Defaults to short_video_cn.'),
   },

package/src/record-url-narration-tool.js

@@ -48,6 +48,19 @@ function derivePhaseCount({ plan, recorderOutput }) {
   return null;
 }
 
+const PIPELINE_SENTINEL_KEY = 'detail_sections_version';
+
+function assertPipelineCompliance(plan) {
+  if (!isPlainObject(plan)) return;
+  if (!plan[PIPELINE_SENTINEL_KEY]) {
+    throw new Error(
+      'pipeline_violation: plan must come from detail_sections output. '
+      + 'Required pipeline: analyze_page → plan_video → detail_sections → record_url_narration → compose_video → submit_to_library. '
+      + 'Do not hand-write phases or bypass detail_sections.'
+    );
+  }
+}
+
 export function validateRecordUrlNarrationArgs(args = {}) {
   const normalizedUrl = normalizeText(args.url);
   if (!normalizedUrl) {
@@ -113,6 +126,12 @@ export async function runRecordUrlNarrationTool({
     return toolError(`Error: ${error.message}`);
   }
 
+  try {
+    assertPipelineCompliance(validatedInput.plan);
+  } catch (error) {
+    return toolError(`Error: ${error.message}`);
+  }
+
   try {
     const result = await runMandatoryLocalToolFn({
       toolName: 'record_url_narration',
@@ -141,7 +160,6 @@ export async function runRecordUrlNarrationTool({
           viewport: finalInput.viewport,
           fps: finalInput.fps,
           settle_ms: finalInput.settle_ms,
-          page_zoom: finalInput.page_zoom,
         });
 
         return {