agentsclimarketplace

Agnes understand video

Skill oh-summy/skills/skills/agnes-understand-video

Cross-agent AI skills for Claude, Codex, OpenCode, Kimi Code, and more. Install all at once or pick only what you need.

Install
npx -y skills add oh-summy/skills --skill agnes-understand-video

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Analyze a video with Agnes AI by sampling frames and return structured information. Requires ffmpeg and ffprobe.

SKILL.md

2.6 KB, 652 tokens by cl100k_base, as published. Nobody here has run it

Agnes Understand Video

Understand a local video by sampling frames and analyzing them with agnes-2.5-flash.

When to use

  • The user asks "what happens in this video", "summarize this video", or "extract key events from this clip".
  • A local video file path or publicly accessible video URL is available.

Inputs

  • --video: Path to a local video file or a publicly accessible video URL.
  • --prompt: Analysis instruction.
  • --timestamp: Analyze around a specific timestamp, e.g. 30, 30.5, 00:00:30, 0:30.
  • --window: Seconds around --timestamp to sample. Default 2.
  • --frames: Number of frames to sample. Default 5.
  • --format: text, structured (default), or json.
  • --json: Alias for --format json.
  • --max-width: Resize frames to this width to save tokens. Default 512.
  • --model: Optional model override. Default is agnes-2.5-flash (or the AGNES_MODEL env var if set).

Steps

  1. Confirm the video source and what the user wants to know.
  2. Run the helper script in this skill directory: 
    bash scripts/understand_video.sh --video ./clip.mp4 --prompt "Summarize the main events in this video"
  3. The script downloads the video if needed, extracts frames, sends them to agnes-2.5-flash, and returns the analysis.

Examples

Analyze the whole video:

bash scripts/understand_video.sh \
  --video ./demo.mp4 \
  --prompt "Describe the setting, characters, and actions in JSON format" \
  --json

Analyze what happens around the 30-second mark:

bash scripts/understand_video.sh \
  --video ./demo.mp4 \
  --timestamp 30 \
  --window 2 \
  --prompt "What happens around the 30-second mark?"

Output format

By default the model returns a structured analysis including:

  • Summary
  • Keyframes analyzed (with timestamps)
  • Scene-by-scene description
  • Settings
  • Characters / actions
  • Camera movement
  • Visual style
  • Mood / tone
  • Audio description
  • Metadata (duration, resolution, frame rate, format)

The script also prints the extracted keyframe timestamps so you know exactly which moments were analyzed.

Use --format json to get machine-readable JSON. See references/output-schema.md for the full schema.

Templates / references

  • references/output-schema.md — structured output schema

API reference

Requirements

  • bash, curl, jq, ffmpeg, ffprobe
  • AGNES_API_KEY set or .env configured

What ships with it: 3 files

9.9 KB alongside SKILL.md, 1 of them executable

references/

scripts/

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.