ai-video-shot-review Skill

Purpose

Claude cannot play video directly, but it CAN read image frames. This skill bridges the gap: take a generated AI video output, extract frames at tactically chosen timestamps with ffmpeg, read them alongside reference start/end frames, and produce a weighted scorecard + production-readiness verdict so the user can decide whether to ship the shot, re-roll, or switch model.

When to Use

Invoke this skill whenever:

• The user has a model output (Kling / Veo / Sora / Runway / Luma / Wan / Seedance / etc.) and wants to evaluate it
• The user mentions phrases like "review this video", "debug this shot", "score this", "analyze this output"
• The user is running an A/B model comparison test and has a new result
• The user wants to decide whether to commit to batch production with the current model

Do NOT invoke for:

• User wants to write a prompt (not reviewing output)
• User wants to plan a shot (nothing generated yet)
• Video is live-action footage (not AI-generated)