The body is information-dense and mostly free of generic concept explanation, but it is a long wall of dense prose contract bullets with heavy inline jargon and repeated caveats that could be tightened. Not 3 because several sentences restate constraints multiple ways (e.g. libass/PIL-overlay fallback described twice) and the token budget carries redundancy; not 1 because it does not explain basics Claude already knows.

It gives concrete specifics (ffmpeg overlay filter with enable='between(t,st,en)', afade tail, loudnorm I=-14, 2 fps contact sheet, FACE/HAND crops) but presents them as prose contract bullets rather than copy-paste executable code blocks, and the full assembly command sequence is absent. Not 3 because there is no executable, complete code/command block a model could run; not 1 because the guidance is specific and concrete, not vague.

The assembly steps are clearly sequenced (cut each clip to bar window, hard-concat, burn logo+captions, append end card, mux song, loudnorm) and there is an explicit per-scene QC validation loop with concrete checks (2 fps contact sheet + per-bar FACE/HAND crops, confirm matte felt + one character + no human hands, re-check served bytes, regenerate KEYFRAME not re-cut on drift). Not 2 because validation checkpoints and the failure-recovery feedback loop are explicit, matching the level-3 anchor.

The body is a concise overview that clearly signals one-level-deep references to real bundle files — 'scripts/config.example.json is the worked example', 'scripts/PIPELINE.md maps every config block', 'scripts/README.md documents the free assembly' — all confirmed to exist. Not 2 because references are well-signaled and one level deep with content appropriately split, not inlined; navigation is easy.

Lists many concrete actions explicitly: cut-to-bar, hard-concat, logo bug + captions burn, solid-color end card, song mux, librosa beat-tracking, VEED-whisper captions — multiple specific concrete assembly actions. Not the level below (which names domain and some actions but is not comprehensive); it exhaustively enumerates the assembly pipeline.

Clearly answers both: what it does ('Assemble a cartoon...music-video ad from a config', 'FREE deterministic assembly stage') and when to use it ('Use for the cartoon-music-video format'). Explicit 'what' AND explicit trigger guidance present, matching the level-3 anchor; not 2 because 'when' is explicit, not merely implied.

Contains relevant domain keywords ('cartoon', 'animated', 'music-video ad', 'config') but most are specialized jargon ('i2v clips', 'look pack', 'Alignment 2', 'bar grid', 'librosa') rather than natural terms a user would actually say when they need this skill. Not 3 because common user-facing variations ('animated video', 'music video', '2D cartoon ad') are thin; not 1 because recognizable media-ad terms do appear.

Highly distinct niche — 'cartoon / animated / hand-crafted music-video ad' assembly with a named format and explicit boundary against the paid create-* skills, making it unlikely to trigger for the wrong skill. Not 2 because the format trigger and scope separation are explicit and specific; it carves a clear niche rather than a generic 'works with videos' overlap.