# Youtube Clipper

> Generate short-form clips from a YouTube video using the Higgsfield personal clipper. Use when the user pastes a YouTube link and asks to "clip it", "make clips", "create shorts", "turn this into clips", or any request to extract short clips/highlights from a YouTube video.

- Skill: `rrijssenbeek/youtube-clipper` (Agent Skill)
- Install (CLI): `npx skillmds@latest add rrijssenbeek/youtube-clipper`
- Raw SKILL.md: https://api.skillmd.com/api/skills/rrijssenbeek/youtube-clipper/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: rrijssenbeek (https://skillmd.com/u/rrijssenbeek)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/rrijssenbeek/youtube-clipper

---


# YouTube Clipper

You turn a YouTube video into short-form clips using the **Higgsfield MCP** — specifically the personal clipper tools (`mcp__claude_ai_Higgsfield__personal_clipper_*`, powered by FNF Clipify). All clipping happens through the Higgsfield MCP server; do not attempt to clip via ffmpeg, yt-dlp, or any other tool.

## Required Inputs

- A YouTube URL (from `$ARGUMENTS` or the user's message). If missing, ask once and stop.
- Default to **10 clips, 9:16 aspect, "Inter" subtitle font** unless the user specified otherwise. Don't re-ask if they already said "10 clips" — just confirm the defaults in one line and proceed.

## Workflow

1. **Load Higgsfield MCP clipper tools** via ToolSearch (the Higgsfield MCP tools are deferred and must be loaded before use):
   ```
   select:mcp__claude_ai_Higgsfield__personal_clipper_create,mcp__claude_ai_Higgsfield__personal_clipper_status,mcp__claude_ai_Higgsfield__personal_clipper_jobs,mcp__claude_ai_Higgsfield__list_workspaces,mcp__claude_ai_Higgsfield__balance
   ```

2. **Sanity check workspace + credits** (parallel): `list_workspaces` + `balance`. If `is_selected` workspace is missing or credits look insufficient, stop and tell the user.

3. **Create the job** with `personal_clipper_create`:
   - `urls`: `[<youtube_url>]` (it's an array)
   - `clips_num`: 10 (max 20)
   - `clip_aspect`: `"9:16"` (also valid: `"1:1"`, `"16:9"`)
   - `subtitle_font`: `"Inter"` (or any of: Noto Sans, Noto Serif, Noto Sans Display, IBM Plex Sans, M PLUS Rounded 1c, Bebas Neue, Archivo Black, Unbounded, Montserrat, Bangers, Permanent Marker, Playfair Display, Caveat)
   - Capture the returned **`row_id`** (UUID) — that's what status polling needs.

4. **Poll `personal_clipper_status`** with `row_id`:
   - Clipping takes several minutes. Use `ScheduleWakeup` — first wake at `delaySeconds: 270` (stays in prompt cache), then 600–1200s for subsequent checks.
   - On failure, surface the error verbatim and stop.

5. **Deliver clips** when complete:
   - Numbered markdown list with: hook/title, duration, any score returned, and the playable/download URL.
   - Report the actual count returned — never claim 10 if fewer came back.

## Critical Rules

1. Never invent a YouTube URL — ask if missing.
2. Always use the **Higgsfield MCP** for clipping — never ffmpeg, yt-dlp, or any other local tool.
3. Always load Higgsfield MCP tools via ToolSearch before calling them.
4. `urls` must be an array, even for a single video.
5. Status polling uses `row_id` (UUID), not "job_id".
6. Don't busy-poll. Use `ScheduleWakeup` at 270s+ intervals.
7. `clips_num` is capped at 20.
8. Echo the actual clip count returned; don't pad.
9. Surface every per-clip field the API returns (hook, score, duration, transcript) — that's the signal users pay for.

## Quick Template — Final Delivery

```
Done — N clips from <video title or URL> (row_id: <uuid>):

1. **<Hook>** — 0:42 · score 87
   https://higgsfield.../clip/...

2. ...
```

Use `$ARGUMENTS` as the YouTube URL when present.