agentleFS
akachandan1akachandan1/video-studio/.claude/skills/product-demo/SKILL.md

product-demo

Skill2 starsChanged 30 days agoMITView on GitHub

Turn a raw screen recording (+ optional timed VO script .srt) into a polished product walkthrough in HyperFrames — automatic cursor/click tracking, camera that eases in to where the user clicks, click ripples, spotlights, chapter cards, stat callouts timed to the voiceover, captions, offline Kokoro VO and a ducked music bed. Works for 1-minute clips up to 10+ minute demos.

13 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 10 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    All later coordinates are in this cropped content space.

    line 15
  2. Very specific:

    Check 5–10 events by pulling the frame at each t and confirming the cursor is where it says.

    line 22
  3. Distinctive:

    No script: write one from the recording (chapter per screen, ≤ 2 lines per action), get it approved, then TTS.

    line 33
  4. Very specific:

    Write demo.json (all fields documented at the top of $T/director.py): content size, duration, brand strings, accent, reading framing …

    line 36

“Common” is measured: kept by at least 6 in 10 of the 10 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Product demo from a screen recording (HyperFrames)

Tools live in .claude/skills/product-demo/tools/ (call it $T). Run them from the project folder videos/<name>/, with the studio venv python (.venv/bin/python, has opencv/numpy/soundfile).

1. Prepare

  1. ffprobe the recording (size, fps, duration, audio?). Pull a frame at 5 s and look at it: find the content crop (drop browser chrome / black bars) as x,y,w,h.
  2. Crop + conform once: ffmpeg -i rec.mp4 -vf crop=w:h:x:y,fps=30 -c:v libx264 -crf 18 -an assets/demo.mp4. All later coordinates are in this cropped content space.

2. Track clicks

python3 $T/track.py assets/demo.mp4 work/track.json          # cursor blobs + UI-change frames (≈ real time ÷ 3)
python3 $T/events.py work/track.json work/events.json        # click / scroll events with cursor position + 'still'

Check 5–10 events by pulling the frame at each t and confirming the cursor is where it says. Clicks with still ≥ 0.25 s (cursor rested before the UI changed) are the reliable ones. If the cursor style differs, a template match around each event (crop the cursor tip from one frame, cv2.matchTemplate, keep score ≥ 0.75) tightens positions.

3. Voiceover (optional)

  • Client gave a timed script (.srt): ~/.local/share/hf-kokoro/bin/python $T/vo_srt.py script.srt vo --voice af_heart --fix pronounce.json — one WAV per block fitted to its slot (speeds up to 1.18× max; reports lines that still overrun). pronounce.json spells acronyms/brands as spoken ({"SSO": "S S O", "SaaS": "sass"}). Listen to 3 lines.
  • Premium voice later: regenerate vo/*.wav + vo/vo.json with ElevenLabs/Higgsfield TTS (quote cost) — nothing else changes.
  • No script: write one from the recording (chapter per screen, ≤ 2 lines per action), get it approved, then TTS.

4. Direct

Write demo.json (all fields documented at the top of $T/director.py): content size, duration, brand strings, accent, reading framing (default view between clicks, e.g. content area with sidebar trimmed at ~1.3×), sidebar_x (navigation clicks get a brief hold), chapters (tied to VO blocks), callouts (big number + label at the moment a VO phrase is said), focus windows for things worth a zoom that have no click (spotlight: true dims the rest).

python3 $T/director.py demo.json                                   # -> index.html + work/director.json
python3 $T/mix.py --vo vo/vo.json --total <TOTAL> --intro 4 --bed music.wav --clicks work/events.json
npx --yes hyperframes@0.8.172 lint && npx --yes hyperframes@0.8.172 check

Music bed: synthesize one with the score skill (calm pad/piano) or use a licensed track. Ducking is automatic.

5. Review & render

Snapshot ~10 moments (each chapter card, 3 zooms, 2 callouts) and look: zoom target actually on screen, callout on the emptier side, captions not covering the action. Tune focus/reading, re-run director. Render (long demos take a while — run in background), loudnorm −14 LUFS, send.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.