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 well-orchestrated skill body: a gated 7-step pipeline with explicit validation checkpoints, feedback loops, and clean delegation of detail to real, well-organized reference files. The only weaknesses are minor redundancy in the autonomous-mode and routing passages and a bare `design-styles.json` mention without a path.
Suggestions
Trim the meta-reasoning paragraph after the non-skippable-gates list ("If you find yourself reasoning... that reasoning is wrong") — the rule it re-explains is already stated explicitly two paragraphs earlier.
Give `design-styles.json` its path (capture/extracted/design-styles.json) at its first mention in Step 1, or defer the mention entirely to step-1-design.md where the path is defined.
The Step 6 sign-off line ("would I post this on social media with my name on it?") could be folded into the existing Gate line as a single clause to save tokens without losing the quality bar.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The step/gate skeleton and Quick Reference tables are dense and earn their tokens, but there are minor instances of over-explanation that could be trimmed: the paragraph "If you find yourself reasoning 'auto mode says bias toward action, so I'll skip X'... that reasoning is wrong" restates the autonomous-mode rule already made two paragraphs earlier, the routing blockquote repeats the frontmatter's NOT-for list nearly verbatim, and "Deliver something you're proud of... would I post this on social media with my name on it?" is motivational padding. It is not a 3 because there is no explanation of concepts Claude already knows and the bulk is efficient; not a 5 because of those few redundant passages. | 4 / 5 |
Actionability | The body provides concrete, executable anchors: `npx hyperframes auth status` with a verbatim-output protocol, the snapshot formula `max(beats × 3, ceil(duration_seconds / 2))`, exact format dimensions, lint/validate zero-error gates, and explicit gate criteria like "`DESIGN.md` exists (any length) with at minimum: color palette, font choices, and do's/don'ts". It is not a 5 because most execution specifics (capture command, TTS pipeline, build patterns) are delegated to reference files, and `design-styles.json` is mentioned without a path, resolvable only inside step-1-design.md; it is not a 3 because the commands, formulas, and gate criteria present are directly executable. | 4 / 5 |
Workflow Clarity | Seven strictly sequenced steps each produce an artifact that gates the next ("Each produces an artifact that gates the next"), every step ends with an explicit **Gate:** checkpoint, user-approval gates (Steps 3–4) are marked, feedback loops exist ("Iterate until they approve", "Fix issues before delivering"), and non-skippable verification checkpoints with checklists (Asset Audit, per-beat HTML read, DoD checklist with per-warning WCAG verification) are enumerated. It is not a 4 because validation is explicit at every step rather than only mostly present. | 5 / 5 |
Progressive Disclosure | The body is a lean overview with clearly signaled one-level-deep references: each step opens with a "**Read:** [references/step-N-*.md]" pointer, and a Reference Files table with a "When to read" column maps every bundled reference to its step; all nine reference files exist, and scripts/w2h-verify.mjs, assets/sfx, and beat-builder-guide.md are properly reachable one hop inside their step files. It is not a 4 because the split is appropriate and navigation is easy — the second-level detail (beat-builder-guide.md) is deliberately and correctly hidden behind step-5-build.md rather than inlined. | 5 / 5 |
Total | 18 / 20 Passed |