Content
71%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body delivers unusually concrete design doctrine — a classification table, exact geometry values, scarcity thresholds, and a clear validation stance — organized into clean sections that properly position the skill on top of its upstream reference. Its one real weakness is redundancy: the overlay law is restated four times across the intro, its section, the rationale section, and the closing bullets.
Suggestions
State the overlay law once in its own section and cut the restatements — the intro paragraph and "Why these two rules are one doctrine" both re-explain it; keep only the non-redundant point that rail (in front) and embed (behind subject) are both layers added to untouched footage.
Merge the four closing bullets ("Keep the full frame... Reserve nothing") into "The overlay law" as a compact summary checklist instead of a separate section, eliminating the fourth repetition.
Give the embed treatment one line of concrete placement guidance (or an explicit pointer to the upstream captions.md section covering matte occlusion / entrance-exit), since it is currently named but never specified.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with real rules (no generic padding Claude already knows), but the overlay law is stated four times: in the intro ("a caption line is composited ON TOP of the film... NOT a reserved zone"), again in "The overlay law" section, again in "Why these two rules are one doctrine", and a fourth time as the closing bullet list ("Reserve nothing... center on true center"). That is more than the minor trimmable repetition of anchor 4, so anchor 3 (mostly efficient but could be tightened) is the best fit. | 3 / 5 |
Actionability | Concrete, executable decision guidance throughout: a drop/rail/embed classification table, hard numbers ("y = H / 2 (landscape 540, portrait 960)", "bottom ~5-8% of canvas height", "bottom ~80px center span"), and crisp thresholds ("≤1 embed per beat, never two adjacent/co-visible, ≥ a beat apart"). Not anchor 5 because embed execution details (matte occlusion, "designed entrance + exit") are named but not specified — they are deferred to the upstream skill without a pointer to where. | 4 / 5 |
Workflow Clarity | The decision rules are unambiguous with explicit if/then coverage ("When captions are disabled: identical positioning freedom"; the pure-cinematic exception "use it only for pure-cinematic asks") and an explicit validation stance ("Snapshot review judges caption-over-content legibility visually" in place of the retired keep-out gate). Not anchor 5 because the overall process — classify phrases, lay out at true center, then visually review legibility — is never sequenced; it must be assembled from separate sections. | 4 / 5 |
Progressive Disclosure | A single self-contained file with well-organized sections (caption model, load-bearing rules, overlay law, rationale) and no nested or buried references. It clearly signals its one-level-deep relationship to the upstream file up front ("supplements the upstream `/hyperframes` skill's `references/captions.md`"), and nothing inlined here clearly belongs in a separate file, which fits the simple-skill guidance that a self-contained, well-sectioned skill can score 5. | 5 / 5 |
Total | 16 / 20 Passed |