Content
63%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 an unusually concrete directive document: real Tailwind classes, specific GSAP parameters, explicit bans, and a mandatory pre-flight validation checklist that gives a clear workflow. Its weaknesses are rhetorical padding that inflates token cost, descriptive (rather than executable) GSAP guidance, and a fully monolithic structure with no progressive disclosure.
Suggestions
Strip the dramatic framing ("catastrophic failure", "BANNED FOREVER", "must breathe") and LLM-behavior commentary down to bare directives to tighten conciseness.
Add a minimal runnable GSAP ScrollTrigger example (setup + pin/scrub config) instead of prose descriptions of the animations, and replace the "simulate a Python script" instruction with a concrete deterministic selection procedure.
Move the catalogs (hero layout options, component arsenal, GSAP paradigms) into a references file (e.g. references/patterns.md) and keep SKILL.md as an overview, adding a post-write verification step to close the workflow's feedback loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The guidance itself is dense and novel (LLM design-failure modes Claude does not inherently know), but it is padded with dramatic emphasis ("catastrophic failure", "BANNED FOREVER", "must breathe") and meta-commentary about LLM laziness that could be trimmed, landing at "mostly efficient with some unnecessary explanation". | 3 / 5 |
Actionability | Highly concrete directives with real class strings ("max-w-5xl", "clamp(3rem, 5vw, 5.5rem)", "grid-flow-dense", "group-hover:scale-105 transition-transform duration-700 ease-out"), a picsum URL pattern, and an embedded code snippet. It falls short of 5 because the GSAP half is descriptive rather than code (no runnable ScrollTrigger setup shown) and Section 1's "simulate a Python script execution" is inherently pseudo-executable. | 4 / 5 |
Workflow Clarity | A genuine pre-flight sequence exists: the mandatory <design_plan> with five numbered verification steps (RNG execution, AIDA check, hero math, bento density, label sweep) followed by "Only output the UI code after this rigorous verification is complete." It misses 5 because validation is pre-code only — there is no post-write check or fix/retry feedback loop, and "Prove mathematically" is under-specified. | 4 / 5 |
Progressive Disclosure | The body is a self-contained ~70-line monolith with zero external files (no references/, scripts/, or assets/ exist). Sections are well-organized, but the catalog material (3 hero layouts, 4 GSAP paradigms, the component arsenal) is reference-like content inlined in the main file, and the under-50-lines well-organized-sections exception does not apply at this length. | 3 / 5 |
Total | 14 / 20 Passed |