@amg555/frameforge-mcp 2.7.1 → 2.7.3

package/plugins/video-pipeline/skills/video-pipeline/assets/project/ANIMATION_GUIDE.md

@@ -156,7 +156,8 @@ Never capture audio from the browser. Audio is a separate pass:
 1. TTS generates audio → word-level timestamps from `scripts/transcribe.py`
 2. Scene resolves animation anchors from those timestamps via `cuePhrase()`
 3. `node scripts/render.mjs full` renders video frames (silent)
-4. `uv run scripts/mix_audio.py` builds the final audio track
+4. `uv run scripts/mix_audio_pro.py --sfx-events --room broadcast --platform youtube` builds the final audio track
+   (reads out/sfx-events.json from window.__events() automatically; use `mix_audio.py` for the simple chain)
 5. `node scripts/render.mjs mux` merges with `-c:v copy` (video stream untouched)
 
 `cuePhrase(phrase)` **throws a fatal build error** if the phrase cannot be found

package/plugins/video-pipeline/skills/video-pipeline/assets/project/pipeline/briefings/orchestrator.md

@@ -34,10 +34,14 @@ Use the right effort level at each stage — wrong effort wastes cost or quality
 7. Run render.mjs sheet (one sheet per shot)
 8. Review sheets, request patches from the responsible subagent
 9. Run render.mjs full (full encode, all workers)
-10. Run verify.py (7-gate QC)
-11. If QC fails: identify the responsible chapter, re-assign to a fresh subagent
-12. Mux audio: render.mjs mux --audio=audio/final_mix.wav
-13. Final verify.py pass
+    → After full render, out/sfx-events.json is auto-generated from window.__events()
+10. Run mix_audio_pro.py (professional signal chain):
+    uv run scripts/mix_audio_pro.py --room broadcast --platform youtube
+    (--sfx-events flag reads out/sfx-events.json automatically if present)
+11. Run verify.py (7-gate QC)
+12. If QC fails: identify the responsible chapter, re-assign to a fresh subagent
+13. Mux audio: render.mjs mux --audio=out/audio_mix_pro.m4a
+14. Final verify.py pass
 ```
 
 ## Subagent rules