Check required CLI tools:
- Run which yt-dlp — if missing: "Install with: pip3 install yt-dlp"
- Run which ffmpeg — if missing: "Install with: brew install ffmpeg (Mac) or sudo apt install ffmpeg (Linux)"
- Run which whisper — if missing: "Install with: pip3 install openai-whisper"
- If ANY are missing, tell the user exactly what to install and stop. All three are required.
Ask:
- "What's your name?" (for CTA sign-offs)
- "Describe your voice style in 2-3 sentences." (e.g. "Direct, casual, no corporate speak. Short punchy sentences. I use real numbers and results.")
- "What's your niche/audience?" (e.g. "business owners who want to automate", "fitness professionals")
- "What's your default CTA?" (e.g. "Comment AUTOMATE for the free guide", "Link in bio for the full breakdown")

Check for a voice file:

Check if ~/.claude/skills/repurpose/voice.md exists

If not, create one from what they just told you:

# Voice Guide

**Style:** [their description]
**Niche:** [their niche]
**Default CTA:** [their CTA]

## Banned Phrases
- "journey"
- "game-changer"
- "in today's world"
- "authentic"
- "leverage your personal brand"

(Add more banned phrases as you use the skill — edit this file anytime.)

Save config:

{
  "name": "User Name",
  "niche": "their niche",
  "defaultCTA": "their default CTA",
  "setupComplete": true,
  "setupDate": "2026-03-26"
}

Say: "Setup complete! Now give me a reel URL and let's repurpose it."

Trigger

User types /repurpose [instagram_reel_url]

Example: /repurpose https://www.instagram.com/reel/DVZrETBDQSw/

Workflow

Step 1 — Ask before doing anything

Before downloading or transcribing, ask this exact question:

"Should I adapt this to your voice, or do you have a specific angle or context you want me to aim for?"

Wait for the answer. Do not download anything yet.

If they say "my voice" or similar → use their voice guide from ~/.claude/skills/repurpose/voice.md
If they give a specific angle or context → note it and use it to guide the rewrite

Step 2 — Download the reel

yt-dlp "[URL]" -o /tmp/repurpose_reel.mp4 --merge-output-format mp4 -q

If download fails → tell the user and stop.

Step 3 — Transcribe

whisper /tmp/repurpose_reel.mp4 --model base --output_format txt --output_dir /tmp/ --fp16 False

Read output from /tmp/repurpose_reel.txt.

Step 4 — Extract the hook

The hook = the first 1–3 sentences (roughly the first 5–10 seconds).

Keep it verbatim or as close as possible. The hook is what stopped the scroll — don't mess with it unless the user asks.

Step 5 — Rewrite the body

Strip everything after the hook. Rewrite in the user's voice OR toward the specified angle/context.

Rules for the rewrite:

Keep the hook verbatim (or near-verbatim)
Same energy and pacing as the original
Strip the original creator's specific branding, tool names only they use, or irrelevant references
Replace with the user's examples, tools, and audience context (from config and voice.md)
Declarative content only — show the WHAT, never teach the HOW (that's the product)
Under 100 words total
End with the user's CTA (from config, or ask if none saved)

Step 6 — Output the script

Present as:

**HOOK** (kept from original)
[hook text]

**BODY**
[rewritten body]

**CTA**
[user's CTA]

Then ask: "Want me to punch up the hook, try a different angle, or save this hook to your database?"

Voice Rules

Read ~/.claude/skills/repurpose/voice.md for the user's voice guide. Follow it strictly. Key defaults:

Direct, casual, no corporate speak
Short sentences. Punchy.
Real numbers and results anchor the body when available
Never teach the HOW in free content — show the WHAT
Avoid all banned phrases listed in voice.md

Hooks Database Integration

After delivering the script, always offer to save the original hook:

"Want me to save that hook to your database? Just say yes and I'll add it."

If yes → extract hook, identify type, append to ~/.claude/skills/repurpose/hooks.md. Create the file if it doesn't exist.

Format:

### [Source Handle] — [Date]
**Hook:** "[the hook text]"
**Type:** [hook type — e.g. curiosity gap, bold claim, pattern interrupt]
**Why it works:** [one sentence]

Cleanup

After transcription, clean up temp files:

rm -f /tmp/repurpose_reel.mp4 /tmp/repurpose_reel.txt

Built by @tenfoldmarc. Follow for daily AI automation builds — real systems, not theory.

repurpose

Repurpose Skill

Purpose

Step 0 — First-Time Setup