Skill: exec-demo-video

Produce a short, high-impact "exec presentation" video (~30–60s) from (a) a live web-app demo link and/or (b) a GitHub repo. This skill adds a targeted upfront questionnaire and exec-cut conventions on top of the general pipeline.

Step 0 — Always load app-demo-video too

This skill does not replace app-demo-video — it layers exec framing on top. Load app-demo-video for the full mechanical detail (bundled Remotion scene components, transition wrapper, assembly template, the generate_tts.py script, and the visual-language / screenshot-capture references). Everything below reuses those bundled assets.

Step 1 — Questionnaire (ALWAYS ask before generating anything)

Work through references/questionnaire.md (voice/language, audience, tone, duration, demo link, repo, optional persona, priority "wow" element, background music, animation/zoom intensity, on-screen emphasis). Never assume language/voice defaults — ask first. Each answer maps to a concrete Remotion/audio setting via the preset table there and app-demo-video/references/animation-and-sound.md. Recommend the standard exec profile (6–7 scenes / 45–55s, dynamic corporate, moderate animation, ducked music) when the user is unsure. Pick a voice from references/voice-catalog.md and offer short samples first.

Step 2 — Script

• One scene = one new idea. Never repeat the same information on screen and in narration — no narration sentence should paraphrase text visible in the shot at that moment.
• Each sentence transitions logically to the next scene.
• Use native vocabulary, never word-for-word translations.
• If a persona is used, keep the on-screen data consistent with it (don't invent a persona that contradicts the repo/demo content).

Step 3 — Captures & animation

• Prefer real screenshots of the app (via browser automation or user-provided) over recreations. Use real documents (PDF, report) as-is: full page first, then zoom to the relevant area.
• Zooms: transform-origin centered on the point of interest (never a pan that pushes content out of frame). Fast ramp to a moderate zoom, then a slow drift for a "living video" feel. Vary the animation pattern between scenes.
• Set zoom strength, entrance snappiness, count-ups (useCountUp), highlight rings (SpotlightRing) and processing bars (LoadingBar) from the intensity answer (Q11) using the presets in app-demo-video/references/animation-and-sound.md.
• If real screen-recordings exist in a different language than the narration, flag it and let the user choose (translated stills vs. real video in the other language's UI).
• Add "processing" elements (progress bars, spinners) synced with narration when the product has real compute time to stage, instead of cutting to the result.
• If the user wants captions (Q13), pass a short caption per scene in the chosen caption language (may differ from the voice) — see app-demo-video/references/animation-and-sound.md.

Step 4 — Music & sound

Set the music from the brief (Q10): if the user provides a track, place it at public/bgm.mp3; if none, remove the BGM <Audio> from the assembly. Keep music ducked (base ~0.05, flares matched to the intensity answer) so it never covers the voice. The bundled MainVideo.template.jsx implements this envelope; tune BGM_BASE/BGM_FLARE per app-demo-video/references/animation-and-sound.md. Never ship copyrighted audio.

Step 5 — Iterative validation

• Render stills (npx remotion still) at key frames BEFORE the full render, for fast user review.
• Only run the full render (npx remotion render) after stills are approved or on explicit request.
• Verify real clip durations (ffprobe) and set each scene's durationInFrames = ceil(seconds*fps) + ~20 accordingly.

Guardrails

• Never use a real third-party/competitor brand as a persona without explicit user confirmation.
• Never show real confidential data — if the demo already uses fictitious/demo data, note it in the video's legal mention if the source product does.
• If the Azure/AAD token for TTS expires (401/tenant mismatch), regenerate it and .strip()/.Trim() the token before using it in an Authorization header (a trailing newline silently breaks the request). See generate_tts.py header.

Expected output

An MP4 in the Remotion project folder (e.g. out_<name>.mp4), plus a short summary of the choices made (voice, duration, scenes) so the user can iterate scene by scene. Produce one language/variant end-to-end before starting another.