مونتاج الفيديو باستخدام Claude Code: 3 برومبتات خطوة بخطوة

You are setting up this folder as my video-editing studio. Explain each step to me in simple Moroccan Darija, and ask before installing anything.

1. Check my OS and what is already installed (node, ffmpeg, uv, hyperframes, whisper-ctranslate2, yt-dlp). Skip what is already there.
2. Install what is missing:
  - Windows (PowerShell):
     winget install OpenJS.NodeJS.LTS
     winget install Gyan.FFmpeg
     powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
  - macOS: brew install node ffmpeg uv (if there is no Homebrew, use the official installers from nodejs.org, ffmpeg and astral.sh/uv)
  Then, on any OS:
     npm install -g hyperframes
     hyperframes browser ensure
     uv tool install whisper-ctranslate2 --python 3.12
     uv tool install yt-dlp
     npx -y skills@latest add heygen-com/hyperframes --all -g -a claude-code -y
     npx -y skills@latest add remotion-dev/skills --all -g -a claude-code -y
  Remotion is not installed globally: it is created inside a project only when a video needs it. Just check that npm can reach it with: npm view remotion version
  If a command is not found right after installing, open a new terminal (PATH) and try again.
3. Test that everything really works:
  - node -v, ffmpeg -version, yt-dlp --version, whisper-ctranslate2 --help
  - make a 3-second test video with HyperFrames (npx hyperframes init _test --example blank --resolution portrait --non-interactive --skip-transcribe, then render it to Exports/test.mp4) and check that the file plays.
4. Create the folders Raw Footage, References and Exports if they are missing.
5. Finish with a short table: tool | version | test OK or not. If something failed, fix it before you stop.
6. Tell me to close this chat and open a new one, so the new skills load.

Learn my editing taste from these videos (3 to 5 links):
1. <link>
2. <link>
3. <link>
4. <link>
5. <link>

1. Download them with yt-dlp into References/. If a link fails, tell me and I will put the file there myself.
2. For each video, measure it with ffmpeg: duration, number of cuts and average shot length (scene detection), and the loudness. Make a contact sheet of its shots and look at it yourself. Transcribe the first 10 seconds with whisper-ctranslate2 to see the hook and the caption rhythm.
3. Note for each one: the hook in the first 2 seconds, the pacing, the caption style (font, size, position, how many words at a time, highlighted words), the colours, the motion between shots, the music energy.
4. Write STYLE.md in this folder: clear, numbered rules my edits must follow (with numbers: shot length, caption size and position, colours as hex codes…), plus what makes these videos work.
5. The key rule: take inspiration from them, not a copy. Never reuse their words, footage, music, logos or characters. My face, my voice and my brand stay mine.
6. Show me STYLE.md in 10 lines of Darija and ask me what to change.

Edit the newest video in Raw Footage/ following STYLE.md. Talk to me in Darija.

1. Transcribe it with whisper-ctranslate2 with word timestamps (my language: Moroccan Darija, use --language ar). Show me the transcript and let me correct the words before you make captions: the captions use my corrected text, not the raw transcription.
2. Cut the dead air: find the silences with ffmpeg (silencedetect) and remove the pauses and false starts, keep all my real sentences.
3. Build the edit as a HyperFrames composition (use the hyperframes skill): 9:16, 1080x1920, word-by-word captions in the style of STYLE.md with the key words highlighted, light motion graphics where they help understand.
4. Music: use the file I put in a Music/ folder, low under my voice. If there is none, ask me (never download music from the internet). Add a few soft sound effects on the visual moments.
5. Before rendering, show me 4 still frames (hook, middle, a caption close-up, end) and a short list of what you cut. Wait for my OK.
6. Render a draft, then the final into Exports/. Check a few frames of the final yourself before telling me it is done.
7. If I ask for a change in plain words ("kbbr l hook", "captions l fou9", "music ghir chwiya"), apply it, show me again, and re-render only what changed.