Content
85%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.
A highly actionable, well-sequenced procedural skill: every step has executable commands, explicit bail gates, and error-recovery fallbacks. The two weaknesses are a monolithic ~470-line structure with no reference files (token-budget tables, sampling reference, and risk tables should be split out) and time-sensitive model-pricing figures inlined in the main body, plus some repetition of the sheet-vs-split rationale across five sections.
Suggestions
Move the Token Budget cost tables, Sampling Method Reference, and Risks and Mitigations sections into a one-level-deep reference file (e.g., references/tuning.md) linked from the body, keeping SKILL.md as a lean pipeline overview.
Replace the time-sensitive per-model figures ('Sonnet 4.6, $3/M', '$0.057') with relative token counts (e.g., '≈ 18.9k image tokens') or move them to a clearly dated reference section, so the skill does not age.
State the sheet-vs-split tradeoff once in the Frame Delivery Mode section and have Step 5d, Step 8, and the Token Budget section reference it, rather than re-deriving the rationale in each place.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Overall lean and assumes competence ('Walk this table top-to-bottom; the first matching rule wins'), but contains trimmable material: time-sensitive pricing/model references ('Cost (Sonnet 4.6, $3/M)', '24 frames at 768 px ≈ 18,864 tokens ≈ $0.057 on Sonnet 4.6') are inlined rather than quarantined, and the sheet-vs-split rationale is repeated across the Frame Delivery Mode, Step 5d, Step 8, Token Budget, and Risks sections. This matches anchor 4 (efficient with minor over-explanation) rather than anchor 5; it is well above anchor 3 because no section explains concepts Claude already knows. | 4 / 5 |
Actionability | Fully executable, copy-paste-ready bash at every stage — ffprobe JSON parsing, keyframe extraction with the correctly escaped filter ('select=\'eq(pict_type\\,I)\''), uniform-fill math, the montage command with the non-obvious mandatory '-frames:v 1', plus concrete fallbacks ('.webm' zero-I-frame fallback, 'scale=768:trunc(ow/a/2)*2') and exact bail/warning messages. Matches anchor 5: executable commands covering the common cases. | 5 / 5 |
Workflow Clarity | Steps 1–9 are strictly sequenced with tool detection mandated 'before any other step', explicit bail-gate table checked 'in order before proceeding', conditional steps with ALL-of conditions, explicit error-recovery paths (keyframe→uniform fallback, sheet→split escalation), and a Risks and Mitigations table. Matches anchor 5 (clear sequence, explicit validation checkpoints, feedback loops for error recovery); the operation is read-only analysis, so no destructive-batch cap applies. | 5 / 5 |
Progressive Disclosure | The body is well-headered and navigable, but it is a ~470-line single file with no bundle files at all (no references/, scripts/, or assets/ exist). Content that clearly belongs in one-level-deep reference files — the Token Budget cost tables, the Sampling Method Reference, and the Risks and Mitigations table — is inlined in SKILL.md, matching anchor 3 ('content that should be separate is inline'). It is not anchor 4 because there are no external references and the main body carries detail that would normally be split out. | 3 / 5 |
Total | 17 / 20 Passed |