[Tejashmakwana]

What

The changelog-video skill now builds the weekly video in the v2 look, taken from the hyperframes.studio launch film. Next Monday's run starts from the new reference build, so this changes the video Rames makes with no other step.

Old and new for the same week (Nov 2-9), side by side: https://www.heygenverse.com/a/969c5b4c-b4a4-4bb2-ba4c-10084fae9bc2

	Before	v2
Look	Dark green ground, the same headline + card every scene	Paper ground and the Studio window on the mint→blue stage, a new framing per scene
Product	Placeholder shapes	The real UI acting out each change (renders stacking at 1/4/8 workers, the motion-blur inspector, the Storyboard tab, a dropped upload part resending, check in a second)
Phone	UI text 10-14 px	Headlines 64-72 px, UI text ≥ 22 px
Cuts	Cross-fades	The clicked thing becomes the next shot
Muted	Caption rail	Headlines carry the meaning
Sound	Annie + music bed	Annie at ~160 wpm (was 178) + music + a sound on every action
Logo	Corner chrome on every frame	End card only

Why

James handed me the Monday changelog to improve and auto-post (Nov 4). The old video was hard to read on a phone and showed placeholder shapes instead of the product. The auto-post half is #7614.

How

• examples/v2/: the Nov 2-9 build (root kit CSS + 7 scene compositions + tools/build_audio.py + that week's BRIEF.md). Every new week copies it and replaces scene content.
• tools/build_audio.py: one place for pacing and sound. HOLDS add pauses inside scenes, VO_SEGS cut the single Annie file at its phrase pauses, and cue(...) puts a sound on each action. It writes every <audio> clip and each scene's start/length into index.html. VO=0 builds a no-voice mix.
• Sound effects come from skills/media-use/audio/assets/sfx/ (credited in its CREDITS.md) and the music is the skill's existing bgm.mp3, so no new third-party audio is added.
• New font weight: TT_Norms_Pro_Light.woff2, the same file the docs already load from www-static-assets.heygen.com.
• SKILL.md rewritten around v2: bootstrap, pick the five, scene plan, the two-layer script and Annie TTS (unchanged), build, gates. The old dark skeleton and build-spec.md stay for reference, marked as not to be mixed.
• examples/v2/compositions/ is force-added because the root .gitignore ignores compositions/ (same as the remotion skill's test corpus).
• .agents/skills/ mirrors .claude/skills/ byte for byte.

Related work

Refs #7614 (auto-post on merge), #7453 (the Nov 2-9 digest this build covers).

Test plan

• Fresh bootstrap exactly as SKILL.md step 0, plus that week's Annie VO: build_audio.py writes 66 clips, hyperframes check 0 errors, render 47.7 s, -14.3 LUFS / -1.0 dBFS peak (the old video is -14.3 LUFS)
• Rendered again after oxfmt: SSIM 1.000 against the first render
• oxfmt --check, check-skill-mirror, check-core-skills pass
• Next Monday's run (Nov 16) through the new skill

🤖 Generated with Claude Code