{
  "markdown": "# ffmpeg-analyse-video-skill\n\nAI agent skill that analyses video content by extracting frames with ffmpeg and using vision to generate timestamped step-by-step summaries.\n\nWorks with screen recordings, tutorials, presentations, footage, and animations.\n\n## Install\n\n```bash\nnpx skills add fabriqaai/ffmpeg-analyse-video-skill\n```\n\n## Requirements\n\n- **ffmpeg** and **ffprobe** installed on your system\n  - macOS: `brew install ffmpeg`\n  - Ubuntu/Debian: `sudo apt install ffmpeg`\n  - Windows: `choco install ffmpeg` or `winget install ffmpeg`\n\n## Usage\n\nProvide a video file path and ask your agent to analyse it:\n\n```\nAnalyse this video: /path/to/recording.mp4\n```\n\n```\nWhat happens in this video? ~/Desktop/demo.mov\n```\n\n```\nSummarise this recording: ./tutorial.mp4\n```\n\n### Advanced\n\n```\nAnalyse 2:00 to 5:00 of meeting.mp4\n```\n\n```\nAnalyse this video in high detail: demo.mp4\n```\n\n```\nFocus on the code shown in this video: screencast.mp4\n```\n\n## How It Works\n\n```\nMain Agent                          Sub-Agents (disposable context)\n──────────                          ──────────────────────────────\n1. ffprobe metadata        ───►\n2. ffmpeg frame extraction ───►\n3. Split frames into batches ──►   4. Read images (vision)\n                                      Write text descriptions\n                                      to batch_N_analysis.md\n5. Read text files only    ◄───    (context discarded)\n6. Synthesise final output\n```\n\nFrame images are only ever read inside disposable sub-agent contexts. The main agent receives lightweight text-only analysis files — no images enter the main conversation. This cuts context usage by ~90% compared to reading frames directly.\n\n### Frame Extraction Strategies\n\n| Video Duration | Strategy | Expected Frames |\n|---------------|----------|-----------------|\n| 0-60s | Interval (1 frame/2s) | 1-30 |\n| 1-10min | Scene detection | 15-60 |\n| 10-30min | Keyframe extraction | 30-80 |\n| 30min+ | Thumbnail filter | Capped at 60 |\n\n### Output Format\n\nThe skill produces a structured markdown report:\n\n- **Metadata** — duration, resolution, fps, content type, frames analysed\n- **Timeline** — chronological segments with descriptions\n- **Key Moments** — 3-7 most significant timestamps\n- **Summary** — 2-5 sentence narrative\n\n## License\n\nMIT\n",
  "bytes": 2284,
  "sha": "b24194fb697ebe7d3cbe14fee8b37c6d80c4c8f8e0de4172245f201a6c34e75d",
  "repo_slug": "fabriqaai/ffmpeg-analyse-video-skill",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/skl_fabriqaai_ffmpeg_analyse_video_skill_ffm_9c88e9dc/readme"
}