The body is lean with efficient tables and code blocks and never over-explains concepts Claude already knows (ffmpeg, Whisper); the only minor inefficiency is the opening sentence restating the description, which is not enough to drop below 3.

It provides copy-paste-ready commands with real script paths, a full CLI options table, and concrete ffmpeg expressions like "select='gt(scene,0.3)'"; not a 2 because the examples are fully executable, not pseudocode.

As a single-task skill, invocation is unambiguous with the three extraction modes clearly compared and the scene-to-interval fallback documented; the simple-skill scoring note allows a 3 without a multi-step workflow.

The body is a concise overview that defers the full schema to a clearly signaled, one-level-deep reference ("See `references/output-format.md`") which is a real file; not a 2 because the reference is well-signaled and not nested.

"Understand video content locally using ffmpeg frame extraction and Whisper transcription" names concrete tools and actions, and the triggers add "Extracting key frames for visual analysis" and "Transcribing video audio locally", listing multiple specific actions.

It explicitly states what ("frame extraction and Whisper transcription") and when via the numbered "Use when:" clause, clearly answering both; not a 2 because the trigger guidance is explicit rather than implied.

Triggers like "Understanding what a video contains", "Transcribing video audio locally", "Extracting key frames", and "video content" are natural phrases a user would say; not a 2 because it covers common variations rather than jargon-only terms.

The local ffmpeg+Whisper niche plus "No API keys needed" gives it a clear, distinct scope unlikely to trigger other skills; not a 2 because it is not merely "works with video files".