skilld
All skills
aktsmm avatar

/video-watch

@91747a7
by yamapanaktsmm/agent-skills26 stars
4

Prepare video URLs or local video files for GitHub Copilot analysis by extracting captions, sampled frames, contact sheets, and a prompt packet. Use when the user asks Copilot to watch, inspect, summarize, or diagnose a video, screen recording, demo, webinar, YouTube/Loom/TikTok/X/Vimeo URL, or .mp4/.mov/.mkv/.webm file.

Ask your Agent

Use this Skill: https://skilld.dev/gh/aktsmm/agent-skills/video-watch

This session only. Nothing lands on disk.

referencesworkflow.md

≈663 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Video Watch Workflow

Use this reference when the user asks for detailed operation or when the first pass is too sparse.

Artifact Contract

The script creates one run directory containing:

Artifact Purpose
manifest.json Machine-readable source, artifact list, warnings, and limitations. Read first.
prompt.md Short task packet for Copilot. Read second.
transcript.md Captions, sidecar transcript, or an explicit unavailable notice.
frame-index.md Frame file names and approximate timestamps.
contact-sheet.jpg Grid image for visual scan.
frames/ Individual sampled frames. Read only selected frames when necessary.

Standard Run

python .github/skills/video-watch/scripts/video_watch.py "<url-or-path>" --question "<question>"

Use an existing transcript from another tool or speech backend:

python .github/skills/video-watch/scripts/video_watch.py "<url-or-path>" --transcript-file "<transcript.txt>" --question "<question>"

Use --start and --end when the user names a moment:

python .github/skills/video-watch/scripts/video_watch.py "<url-or-path>" --start 2:15 --end 2:45 --question "what changes here?"

Copilot Consumption Pattern

  1. Read manifest.json and note warnings.
  2. Read prompt.md to preserve the user's question.
  3. Read transcript.md for spoken content.
  4. Read frame-index.md and inspect contact-sheet.jpg.
  5. Inspect individual frames only when the contact sheet points to relevant timestamps.
  6. Answer with evidence boundaries: transcript-only, frame-only, sparse sampling, or missing captions.

Review Checkpoints

Use a producer/critic loop when changing this skill:

  1. Producer drafts the plan or implementation.
  2. Critic reviews read-only for primitive choice, scope, artifact order, safety boundary, dependency boundary, self-contained resources, and validation.
  3. Producer fixes blocking findings and reruns the smallest relevant validation.
  4. Repeat until PASS or PASS_WITH_NOTES, with a maximum of 2 critic rounds per checkpoint unless the user asks for more.

Checkpoint targets:

  • Plan review: before adding or changing workflow behavior.
  • Implementation review: after editing SKILL.md, scripts, or references.

Source: SKILL.md on GitHub

Third-party checks

1 alert2mo3 checks · Risk SAFE
  • Gen Agent Trust Hub2mo

    The skill extracts artifacts from video files and URLs for analysis by an AI agent. It uses standard command-line tools like ffmpeg and yt-dlp for processing. The skill implements safety measures such as redacting sensitive URL parameters and providing user warnings for potential credential exposure in URLs.

  • Socket2mo

    No alerts

  • Snyk2mo

    Risk: CRITICAL · 2 issues

Provenance

Signed by skilld at 91747a7. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 18 hours ago.

Activeupdated last week

Capability

argument-hint
動画URLまたはローカル動画パス、質問、必要なら時間範囲
user-invocable
true
metadata
{
  "author": "yamapan (https://github.com/aktsmm)"
}

README badge

README badge for aktsmm/agent-skills/video-watch