Content
80%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 dense, disciplined template spec: exact colors, grid, typography, and a fixed 22-layout pool with hard constraints, all token-efficient and easy to navigate. Its main weakness is workflow — the body states the rules but never sequences the build (clarify inputs → select theme → map content to layouts → emit → critique) and surfaces no validation step, relying on the frontmatter example prompt for the critic pass. Adding a short numbered build sequence with the 'board member knows what to approve before page five' check would complete it.
Suggestions
Add a short numbered workflow in the body (clarify missing essentials → choose one of the 4 themes → map user content onto the layout pool until fully covered → apply the design iron rules → output single-file HTML), instead of leaving the sequence implicit in section order.
Surface the validation/critic pass in the body (e.g. a final check step: 'Critic pass: would a board member know what to approve and why before page five?') so the quality checkpoint is part of the skill, not just the frontmatter example prompt.
Include one minimal executable snippet (e.g. the 16-column grid + hairline card skeleton in HTML/CSS) to make the layout guidance copy-paste ready rather than inference-based.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every line is a load-bearing rule — hex values ("accent `#002FA7`, paper `#fafaf8`, ink `#0a0a0a"), concrete CSS ("`grid-template-columns: repeat(16, 1fr); gap: 0`"), font stack, and hard constraints ("只用直角: 全程 `border-radius: 0`"). Nothing explains concepts Claude already knows and there is no padding; this matches anchor 5 (lean, every token earns its place). | 5 / 5 |
Actionability | The spec is highly concrete — four locked themes with exact hexes, 22 named layouts, typography and size rules, keyboard/hash navigation, and the output contract ("输出单文件 HTML, 不用任何外部图片 URL") — so a builder knows exactly what to produce. It falls at anchor 4 rather than 5 because there is no executable HTML skeleton or worked example of any single layout, and one-line layout descriptions (e.g. "S08 Duo Compare — 垂直分割线; 左 Before / 右 After") leave implementation details to inference. | 4 / 5 |
Workflow Clarity | The sections imply a rough order (pick theme → cover content with layouts from the pool → apply design rules → emit single-file HTML), but no explicit step sequence is given and there is no validation or critic pass in the body — the quality check ("would a board member know what to approve and why before page five") appears only in the frontmatter example_prompt, not the body. This matches anchor 3 (sequence present implicitly, checkpoints missing) and is capped there by the absent validation loop. | 3 / 5 |
Progressive Disclosure | The body is ~43 lines with no bundle files (no references/, scripts/, or assets/ exist) and is cleanly sectioned into 【意图】/【主题】/【布局】/【设计细节】 with the 22 layouts as a scannable list. Per the rubric's simple-skill exception (under 50 lines, no need for external references, well-organized sections), this matches anchor 5. | 5 / 5 |
Total | 17 / 20 Passed |