landing-audit

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Landing Audit

着陆页审核

Grade a landing page and hand back a scored, prioritized to-do list. The output is findings + a punch-list, not edits — only change code if the user asks afterward.

Two rubrics, merged:

Scored core = Marc Lou's 31 viral-product principles (verbatim in
```
31-principles.md
```
) → one reproducible number out of 100.
Unscored lens = the design-rules rulebook (UI craft & feel: hierarchy, type, color, spacing, calm, micro-interactions, motion, load speed) — invoke the
```
design-rules
```
skill and grade the rendered page against it, reported alongside but kept out of the number. That skill is the single source of truth; this audit applies it, it does not restate it.

Detection tiers, the scoring formula, categories, and the report shape all live in

grading.md

. Read it before grading.

对着陆页进行评级，并返回带评分的、按优先级排序的待办事项清单。输出内容为审核结果+待办清单，而非直接编辑——仅在用户后续要求时才修改代码。

采用两套合并的评估标准：

核心评分项 = Marc Lou的31个病毒式产品原则（原文详见
```
31-principles.md
```
）→ 得出100分制的可复现分数。
非评分视角 = 设计规则手册（UI工艺与质感：层级、字体、色彩、间距、视觉舒适度、微交互、动效、加载速度）——调用
```
design-rules
```
技能，对照渲染后的页面进行评级，结果将与核心评分一同展示，但不计入总分。该技能为唯一权威依据，本次审核仅应用其规则，不重述内容。

检测层级、评分公式、类别划分及报告格式均详见

grading.md

。开始评级前请先阅读该文档。

Workflow

工作流程

Resolve the target. A live URL, or the project's homepage (Next.js App Router:
```
app/page.tsx
```
and the components it renders; else the framework's index route). For a local app, get it running — prefer the project's own launch path (a
```
run
```
skill or documented dev command, e.g.
```
bun dev
```
->
```
localhost:3000
```
); reuse an already-running dev server.
Render it. Screenshots are non-negotiable (principles 2, 4, 5, 6, 10, 15, 20 and the whole craft lens need them). Use Playwright to capture above-the-fold and full-page at desktop (1440x900) and mobile (390x844), and in both light and dark when the page offers a theme toggle (grade both; a page that only holds up in one theme is itself a finding). Save to the project's own audit/screenshots folder if it has one, else a temp dir — never the project root.
Extract signals. Hero headline, all CTA texts, pricing section, footer, testimonial blocks, meta tags (
```
og:image
```
/
```
twitter:image
```
), color clusters, reading level, hedge-word hits, media embeds. For local apps, cite
```
file:line
```
.
Grade all 31 against the verbatim principles, using the detection tier for each (see
```
grading.md
```
). Every principle gets exactly one verdict — pass / needs-work / fail — counts must sum to 31. Run the craft lens separately.
Score + report. Compute the score with the flat rubric in
```
grading.md
```
(pass=1, needs-work=0.5, fail=0, over 31; one decimal). Emit the full report shape from
```
grading.md
```
: verdict sentence, six category meters, the count line, punch-list (top 3-5), needs-work, passed, the unscored craft lens, and 3 rewrites.

确定目标页面：可以是在线URL，或项目的首页（Next.js App Router：
```
app/page.tsx
```
及其渲染的组件；其他框架则对应索引路由）。对于本地应用，需启动项目——优先使用项目自带的启动方式（
```
run
```
技能或文档记录的开发命令，例如
```
bun dev
```
->
```
localhost:3000
```
）；若已有运行中的开发服务器，可直接复用。
渲染页面：必须生成截图（原则2、4、5、6、10、15、20以及整个工艺视角都需要截图）。使用Playwright捕获首屏和全页的截图，分别对应桌面端（1440x900）和移动端（390x844）；若页面支持主题切换，需同时捕获浅色和深色模式的截图（两种模式均需评级；仅在一种模式下表现良好的页面本身即为问题点）。若项目已有audit/screenshots文件夹，将截图保存至该文件夹，否则保存至临时目录——切勿保存至项目根目录。
提取关键信息：主标题、所有CTA文案、定价板块、页脚、推荐语模块、元标签（
```
og:image
```
/
```
twitter:image
```
）、色彩组合、阅读难度、模糊措辞出现次数、媒体嵌入内容。对于本地应用，需标注
```
文件:行号
```
。
全面评级：对照原文原则对全部31项进行评级，每项采用对应的检测层级（详见
```
grading.md
```
）。每项原则仅给出一个判定结果——通过/需改进/不通过，判定结果总数需为31项。单独运行工艺视角的评估。
计算分数并生成报告：使用
```
grading.md
```
中的统一评分标准计算分数（通过=1分，需改进=0.5分，不通过=0分，总分31分，保留一位小数）。按照
```
grading.md
```
中的完整报告格式输出：判定语句、六个类别指标、统计行、待办清单（前3-5项）、需改进项、通过项、非评分的工艺视角评估结果，以及3份文案改写版本。

Rules

规则

Audit, don't edit. Findings + punch-list only; implement only if asked after.
Every verdict traces to
31-principles.md
— Marc Lou's exact words, never a paraphrase. Quote his language where it sharpens the call.
The score is computed, never hand-typed, and the counts sum to 31.
Precision over coverage — a check you can't run reliably is needs-work / low-confidence, never a wrong confident verdict.
Always look at the rendered page (desktop + mobile, and light + dark when the page is themed); never grade from source alone.
Be specific and cite evidence (
```
file:line
```
, quoted copy, element counts) — no generic design platitudes.
Offer to turn the punch-list into Todoist tasks or to start on the top items.

仅审核，不编辑：仅输出审核结果和待办清单；仅在用户后续要求时才执行修改。
所有判定均需追溯至
31-principles.md
——必须引用Marc Lou的原文，不得自行转述。在需要强化结论时，可直接引用其表述。
分数需通过计算得出，不得手动输入，且判定结果总数需为31项。
精准优先于全面——若某项检查无法可靠执行，判定为需改进/低置信度，不得给出错误的高置信度判定。
必须查看渲染后的页面（桌面端+移动端，若支持主题则需浅色+深色模式）；不得仅通过源码进行评级。
内容需具体，并提供证据（
```
文件:行号
```
、引用文案、元素数量）——不得使用通用的设计套话。
可主动提出将待办清单转换为Todoist任务，或着手处理优先级最高的事项。