@lightcone-ai/daemon 0.16.1 → 0.16.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lightcone-ai/daemon",
-  "version": "0.16.1",
+  "version": "0.16.2",
   "type": "module",
   "main": "src/index.js",
   "bin": {

package/src/chat-bridge.js

@@ -1414,6 +1414,13 @@ server.tool('synthesize_tts',
 );
 
 // ── plan_video_segments ────────────────────────────────────────────────────────
+// Session-scoped flag set when plan_video_segments runs. compose_video_v2
+// refuses TTS-bearing segments unless this is true — the agent must route
+// audio through plan_video_segments first so durations / subtitle_text /
+// audio_path are mechanically aligned. This is a per-chat-bridge-process
+// flag, so a fresh codex session must call plan_video_segments fresh.
+let _planVideoSegmentsCalledThisSession = false;
+
 server.tool('plan_video_segments',
   'Universal audio-video sync planning step. For each segment, call TTS to get the real audio duration, then compute the visual duration with a safety buffer. Returns a planned segments array ready to pass directly to compose_video_v2 (with audio_path, presentation.duration/per_card_duration, and subtitle_text pre-filled). Always call this before compose_video_v2 when you have narration text.',
   {
@@ -1430,7 +1437,11 @@ server.tool('plan_video_segments',
     voice_id: z.string().optional().describe('TTS voice ID. Omit to use workspace default.'),
     workspace_id: z.string().optional().describe('Target workspace. Defaults to current workspace context.'),
   },
-  async (args) => runPlanVideoSegmentsTool({ ...args, currentWorkspaceId, api })
+  async (args) => {
+    const result = await runPlanVideoSegmentsTool({ ...args, currentWorkspaceId, api });
+    if (!result?.isError) _planVideoSegmentsCalledThisSession = true;
+    return result;
+  }
 );
 
 // ── compose_video_v2 ───────────────────────────────────────────────────────────
@@ -1454,7 +1465,35 @@ server.tool('compose_video_v2',
     resolution: z.string().optional().describe('Output resolution WxH. Default "1080x1920".'),
     output_path: z.string().optional().describe('Absolute output path for the mp4. Auto-generated if omitted.'),
   },
-  async (args) => runComposeVideoV2Tool({ ...args, workspaceDir: WORKSPACE_DIR })
+  async (args) => {
+    // Tool-level enforcement of the synthesize_tts → plan_video_segments →
+    // compose_video_v2 standard chain. If any segment has audio_path (i.e.
+    // narration is involved) and the agent never invoked plan_video_segments
+    // in this session, refuse the compose — manual dwell/duration math is
+    // unreliable (last syllable cut, silent tails, subtitle drift). Observed
+    // twice in row: agent skipped plan_video_segments, manually estimated
+    // dwell_ms wrong, ended up with too-long records and silent tails it then
+    // re-recorded to fix — wasting record_url_narration runs that
+    // plan_video_segments would have prevented.
+    const segments = Array.isArray(args?.segments) ? args.segments : [];
+    const hasNarration = segments.some(s => typeof s?.audio_path === 'string' && s.audio_path.trim());
+    if (hasNarration && !_planVideoSegmentsCalledThisSession) {
+      return {
+        isError: true,
+        content: [{
+          type: 'text',
+          text: 'compose_video_v2 refused: TTS-bearing segments (audio_path present) require plan_video_segments '
+            + 'to have run earlier in this session — it mechanically aligns audio_duration / video_duration / '
+            + 'subtitle_text with a safety buffer. Manual dwell/duration math has repeatedly produced misaligned '
+            + 'subtitles and silent tails that force re-recording.\n\n'
+            + 'Standard chain: synthesize_tts(per segment) → plan_video_segments(with text+visual_kind+visual_path) '
+            + '→ compose_video_v2(use the returned segments verbatim, only swap visual_path/visual_kind for real '
+            + 'media). Call plan_video_segments now and pass its output here.',
+        }],
+      };
+    }
+    return runComposeVideoV2Tool({ ...args, workspaceDir: WORKSPACE_DIR });
+  }
 );
 
 // ── take_page_screenshot ───────────────────────────────────────────────────────