Video Highlight Skill

Sponsor · Step Plan

AI video highlight skill for agents: analyze videos, cut clips with FFmpeg, export subtitles, and generate a shareable recap page.

Claude Code · OpenClaw · Codex · Agent Skill

Quick Start

Copy this and send to your AI Agent:

Read this page and install this skill: [inhai-wiki/video-highlight-skill.git](https://github.com/inhai-wiki/video-highlight-skill.git)

Install the complete repository, including SKILL.md, scripts/, references/, assets/, and agents/.

Then give your Agent a video:

Use video-highlight-skill to analyze this video, cut highlight clips, and generate a shareable recap page.

Compatible with Agent frameworks that support GitHub Skill installation, including Claude Code, OpenClaw, Codex, and custom Agent systems.

What It Does

• Analyzes videos and builds timestamped highlight plans
• Cuts highlight clips with ffmpeg
• Exports sidecar SRT subtitles
• Generates a YouTube-style HTML recap page
• Supports long videos, meetings, courses, livestreams, demos, and short-form product videos

Demo

Example: a 20-second product video is split into 3 highlight clips.

Original input video

_{20-second source video used as the Agent input.}

Highlight clips

1. Skepticism turns into curiosity 0:00-0:05.2 · Hook

2. Benefit stack in one beat 0:05.2-0:12.4 · Product value

3. Taste proof and direct recommendation 0:12.4-0:20.1 · CTA

Example plan:

examples/poppi-raspberry-rose-plan.json

Output Page

The generated recap page includes:

• Main video player on the left
• Scrollable highlight playlist on the right
• Video previews for each highlight
• Clip summary, quote, reason, score, tags, and takeaways
• GitHub link in the top navigation and footer

Agent Output Contract

The Agent should generate a clip_plan.json:

{
  "scenario": "highlight",
  "source_title": "Video title",
  "summary": "Short summary.",
  "segments": [],
  "highlights": [
    {
      "start": 0,
      "end": 30,
      "title": "Highlight title",
      "summary": "What happens in this clip.",
      "reason": "Why this clip matters.",
      "score": 90,
      "tags": ["demo"],
      "quote": "Representative quote.",
      "takeaways": ["Key takeaway"],
      "subtitles": []
    }
  ]
}

Full schema: references/analysis-schema.md

Requirements

• Python 3.9+
• ffmpeg
• ffprobe

Speech transcription and multimodal understanding can be provided by your Agent framework.

License

MIT