Converts video tutorials into production-quality SKILL.md files where Claude executes the work, not teaches it.
Give it a YouTube URL, a local video file, or a pasted transcript. It extracts the content, assesses quality AND executability, fills gaps from official docs, verifies that tool actions actually exist before referencing them, and outputs an execution-mode SKILL.md where every step is tagged as EXECUTE (Claude does it), HAND OFF (user's hands required), or DECIDE (user chooses, Claude acts).
| Stage | What happens |
|---|---|
| 1. Extract | Pulls transcript via yt-dlp, Whisper, or pasted text. Detects BOTH the converter environment AND the target environment where the output skill will run. |
| 2. Assess | Scores transcript on completeness, specificity, accuracy, AND executability. Warns if the tutorial is mostly GUI-based and the resulting skill will be hand-off heavy. |
| 3. Research | Fills gaps from official docs. Checks for existing skills. Verifies that every MCP/API action referenced actually exists before writing EXECUTE steps. |
| 4. Write | Generates execution-mode SKILL.md. Every step tagged [EXECUTE], [HAND OFF], or [DECIDE]. Includes metadata (time estimate, step breakdown, dependencies, version info). |
| 5. Self-Check | 20-point validation covering format, sources, tool verification, and execution-mode enforcement. |
| 6. Dry Run | Simulates execution step by step. In Claude Code, actually runs code steps to verify. Catches failures before packaging. |
| 7. Package | Generates README with version notes, skill profile, and licence warning. |
- Execution-first — skills make Claude do the work. Every step classified as EXECUTE/HAND OFF/DECIDE. Claude writes all copy, generates all content, and only asks the user for judgment calls.
- Target environment awareness — asks where the skill will run (Claude Code, Claude.ai with connectors, text-only chat) and adjusts the execute/hand-off split accordingly.
- Tool action verification — before writing "Search Canva for templates" as an EXECUTE step, verifies the Canva MCP actually exposes that action. Prevents skills from breaking on first run.
- Executability scoring — warns upfront if a tutorial is mostly GUI-based and the resulting skill will require heavy manual action.
- Dry run — simulates execution before packaging. In Claude Code, actually runs code steps to catch real errors.
- Version tracking — notes which tool versions the skill was built against and flags when regeneration may be needed.
- Session persistence — every skill includes a "When the user returns" section with a config file strategy so repeat use skips setup.
- Duplicate detection — checks GitHub and SkillsMP before generating.
- Graceful degradation — tool-connected skills include fallback instructions for when connectors aren't available.
- Edit mode — update specific steps without regenerating the entire skill.
cp -r skills/video-to-skill ~/.claude/skills/Copy the SKILL.md file to your tool's skills directory.
Copy the contents of SKILL.md and paste it at the start of a conversation.
Turn this video into a skill: https://www.youtube.com/watch?v=VIDEO_ID
Here's a transcript from a tutorial. Convert it into a SKILL.md file.
Step 4 is wrong — the menu is under Settings > Advanced > API Keys.
Only needed for automated extraction in Claude Code. The skill works without them via pasted transcripts.
- yt-dlp — YouTube transcript extraction (
pip install yt-dlp) - ffmpeg — frame extraction from local video files (opt-in)
- Whisper — transcribing videos without subtitles (
pip install openai-whisper) - webvtt-py — VTT subtitle parsing (
pip install webvtt-py)
- anti-ai-slop-writing — Forces AI to produce human-sounding text (91 stars)
- full-stack-audit — 90-point website audit skill
Created by Jalaaldeen — builder of Wardex, ZakatChain, and open-source AI tooling for founders.
MIT