Content
77%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 is a highly actionable, well-sequenced workflow with concrete commands, CSS invariants, and explicit validation checkpoints throughout. Its two weaknesses are padding (repeated anti-'AI slop' advice and Vercel signup hand-holding) and a broken reference surface: four of the seven referenced supporting files do not exist in the bundle, leaving the core generation phases without their promised materials.
Suggestions
Consolidate the anti-generic-design guidance into the single Design Aesthetics section and cut the repetitions in Core Principles and the closing exhortation ('it is critical that you think outside the box!').
Restore or add the four missing referenced files (STYLE_PRESETS.md, viewport-base.css, html-template.md, animation-patterns.md) — or, if they are intentionally unavailable, inline minimal substitutes — since Phase 3 mandates reading viewport-base.css and html-template.md.
Trim the Vercel first-time-user walkthrough to its decision points (check `npx vercel --version`, `npx vercel whoami`, deploy, share URL); Claude can improvise signup dialogue without a scripted step-by-step.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The bulk is dense, prescriptive spec (viewport rules, density-limit table, exact breakpoints), but there is padding: the anti-generic-aesthetic advice is repeated roughly three times ("No generic 'AI slop.'", the "Avoid generic AI-generated aesthetics" list, and "You still tend to converge on common choices... it is critical that you think outside the box!"), and the Vercel section hand-holds through signup ("Vercel is a free hosting service. You need an account to deploy") plus long prose gotchas. Not 2: most sections earn their tokens with skill-specific rules; not 4: the repetition and hand-holding are noticeable enough to warrant trimming. | 3 / 5 |
Actionability | Nearly every step is copy-paste ready: "python scripts/extract-pptx.py <input.pptx> <output_dir>", "bash scripts/deploy.sh <path-to-presentation>", "npx vercel whoami", "npx playwright install chromium", exact CSS invariants ("height: 100vh; height: 100dvh; overflow: hidden;"), and fully specified AskUserQuestion options. Not 4: no minor gaps in the written guidance — commands, placeholders, fallbacks ("--compact" flag), and recovery paths all cover the common cases. | 5 / 5 |
Workflow Clarity | Clear phase sequence (0-6) with explicit mode detection up front, explicit verification checkpoints ("After ANY modification, verify: `.slide` has `overflow: hidden`, new elements use `clamp()`..."; "Before deploying, verify: open the deployed URL and check that all images load"), and feedback loops for failure recovery (Playwright install failure steps; >10MB PDF → ask user → re-run with --compact). Not 4: validation is explicit and repeated at every risky boundary, including a post-modification checklist for Mode C. | 5 / 5 |
Progressive Disclosure | The structure is well designed — one-level-deep references, a Supporting Files table with a "When to Read" column, and phase-gated reads — but 4 of the 7 referenced files (STYLE_PRESETS.md, viewport-base.css, html-template.md, animation-patterns.md) are missing from the bundle, so mandatory instructions like "read viewport-base.css and include its full contents in every presentation" are unresolvable. Not 4/5: against the actual bundle, navigation to the core generation materials is broken; not 2: SKILL.md itself is a genuine overview with clearly signaled references, not an inlined monolith. | 3 / 5 |
Total | 16 / 20 Passed |