podcast-clipper-subscut（播客 CLIpper 子剪切）

Generate Podcast Clips

Use this skill to convert a long-form spoken video into multiple short clips through the Subscut /podcast-to-clips API.

What This Skill Does

The skill is an opinionated wrapper around the API outcome:

• extracts up to 20 strong short clips from a long-form video
• favors viral and high-retention spoken moments
• adds captions with selectable styles
• supports two render formats: dynamic (auto-reframing) and hook_frame (original frame + title card)
• returns titles, scores, and rendered clip URLs

Think in outcomes, not transport:

• Good framing: "Extract viral short-form content from this podcast"
• Bad framing: "Call some video API"

When To Use

Use this skill when:

• the input is a long podcast, interview, webinar, or talking-head video
• the user wants growth, repurposing, shorts, reels, or TikTok content
• the user wants minimal manual editing

Avoid this skill when:

• the source is already short-form
• the content is mostly non-speech
• the user wants manual, frame-by-frame editing decisions

Do not use it as a generic video editing tool.

Input Contract

Use this compact input shape when planning or explaining the tool call:

{
  "video_url": "https://example.com/video.mp4",
  "max_clips": 5,
  "style": "viral",
  "format": "dynamic",
  "captions": true,
  "clip_duration": {
    "min": 20,
    "max": 60
  }
}

Field Reference

Caption Styles (style)

Render Formats (format)

Use hook_frame when the video is already vertical or the user wants the title displayed prominently. Use dynamic (default) for horizontal/landscape podcasts with one or two speakers.

Output Contract

Expect JSON in this shape:

{
  "clips": [
    {
      "video_url": "https://...",
      "title": "Why Most Founders Get This Wrong",
      "score": 0.92,
      "start": 142.5,
      "end": 198.3
    }
  ]
}

score is a 0–1 float representing clip virality confidence. Higher is better.

CLI Entry Point

Once the skill is installed, the agent runtime should invoke the bundled CLI wrapper:

npm --prefix .agents/skills/generate-podcast-clips run generate-podcast-clips -- \
  --video-url "https://example.com/podcast.mp4" \
  --max-clips 5 \
  --clip-style viral \
  --format dynamic \
  --captions true \
  --min-clip-duration 20 \
  --max-clip-duration 60

Required environment variables:

• SUBSCUT_API_KEY or --api-key

Default base URL:

• https://subscut.com

Install Model

This skill is meant to be published once by Subscut and then installed by users through the marketplace.

End-user flow:

1. Install the published skill from ClawHub / OpenClaw.
2. Set SUBSCUT_API_KEY.
3. Let the agent call the skill when it needs to turn a long-form spoken video

into short clips.

The skill should not ask users to publish or package anything themselves.

Agent Workflow

1. Confirm the source is a long-form spoken video.
2. Prefer the CLI wrapper over hand-written curl.
3. Keep the request simple unless the user asks for custom clip counts, durations, style, or format.
4. Default format to dynamic unless the user explicitly wants a title card (hook_frame).
5. Return the resulting clip URLs, titles, scores, and timestamps.
6. If the API fails, surface the status and response body clearly.
7. Keep a human in the loop before final publishing if the downstream workflow is public.

Natural-Language Triggers

Likely user intents that should map to this skill:

• "Turn this podcast into reels"
• "Make shorts from this interview"
• "Repurpose my long video into clips"
• "Find the viral moments in this podcast"
• "Generate YouTube Shorts from this episode"
• "Make clips with a title card at the top"
• "Give me clean subtitle clips"
• "Extract 10 clips from this webinar"

Notes

• The API route is POST /podcast-to-clips.
• style maps to the API style field. clip-style is used as the CLI flag name.
• format controls render layout: dynamic (default) or hook_frame.
• clip_duration.min and clip_duration.max let callers control clip length window.
• Short videos (≤3 min) are automatically capped at 2 clips regardless of max_clips.
• The script is executed through the skill-local npm script.
• The skill is intentionally opinionated and keeps the parameter surface small for better agent usage.
• This package is for installation and runtime usage, not for publisher-only deployment steps.