agentleFS
grw-labgrw-lab/skills/video-reader/SKILL.md

video-reader

Skill1 starsChanged 35 days agoMITView on GitHub

Analyze any uploaded video (mp4, mov, webm, m4v) by turning it into a contact sheet, key frames, cut timestamps and audio rhythm data, then writing a structured breakdown of scenes, motion techniques, transitions, pacing, colours and typography. Use this skill whenever the user uploads or links a video and asks to review it, break it down, analyse its style, copy, recreate or adapt its look, describe what happens in it, or use it as a reference — even if they just say "look at this video" or "make something like this".

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 1 is distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read <workdir>/summary.json first (duration, size, fps, audio, cuts, onsets).

    line 25
  2. Very specific:

    If something is unclear at a specific moment, extract one more frame: python scripts/analyze_video.py <video> --out <workdir> --frame …

    line 28
  3. Very specific:

    Write the analysis using references/report-template.md.

    line 30
  4. Very specific:

    If onset_regular is false, report median_interval and say the onsets are irregular.

    line 37
  5. Very specific:

    If cuts is empty or the video is one short segment, it is one continuous shot or it uses soft transitions.

    line 38
  6. Very specific:

    Do not read typography, logos or small UI from the contact sheet.

    line 39
  7. Distinctive:

    Treat the reference as style input.

    line 40
  8. Very specific:

    If warnings is not empty, include them in the Limits section.

    line 42
  9. Very specific:

    Every timestamp in the report appears in summary.json, on the contact sheet, or on a frame you extracted.

    line 46

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Video Reader

You cannot watch video or hear audio. This skill converts the video into things you can read: images (a contact sheet and key frames) and numbers (metadata, cuts, loudness, onsets). Never claim to have watched or listened to it.

When to use

  • "Look at this video."
  • "Break this down."
  • "What is the pacing, and where are the cuts?"
  • "Make something like this" or "adapt this style to our brand."

When not to use

  • The input is a still image, a storyboard, or a written shot list. Read those directly.
  • The user wants a transcript of speech or lyrics. Say the audio is numbers only, and ask them to paste the words.

Steps

  1. Run: python scripts/analyze_video.py <video> --out <workdir>
  2. Read <workdir>/summary.json first (duration, size, fps, audio, cuts, onsets).
  3. View <workdir>/contact_sheet.png. This is the whole video at a glance. Each tile has a timestamp.
  4. View <workdir>/keyframes.png. This is the only place to judge typography, UI, lighting and texture.
  5. If something is unclear at a specific moment, extract one more frame: python scripts/analyze_video.py <video> --out <workdir> --frame <seconds>
  6. Write the analysis using references/report-template.md.

The script needs ffmpeg, ffprobe, numpy and Pillow. It writes only inside --out.

Rules

  • Audio is numbers only. You can state loudness shape, onset times, and a BPM only when audio.onset_regular is true. You cannot identify instruments, genre, lyrics or speech. Say so if asked, and ask the user to describe it.
  • If onset_regular is false, report median_interval and say the onsets are irregular. Do not turn that interval into a tempo.
  • If cuts is empty or the video is one short segment, it is one continuous shot or it uses soft transitions. Say that. Do not invent cuts.
  • Do not read typography, logos or small UI from the contact sheet. Tiles are small. If it is not legible in keyframes.png, write "not visible".
  • Treat the reference as style input. When the user wants to recreate it, list what to take and what not to copy (assets, logos, other brands' UI, copyrighted music).
  • Timestamps in the report must come from summary.json or from frames you actually viewed.
  • If warnings is not empty, include them in the Limits section.

Before you finish

  • Every timestamp in the report appears in summary.json, on the contact sheet, or on a frame you extracted.
  • Typography claims cite a key frame, or say "not visible".
  • The Limits section says what the frames and numbers could not determine.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.