AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Talk Visual Review

skill-webkubor-talk-skills-talk-visual-review · by webkubor

演讲页/幻灯片的视觉验收——用浏览器实测每一屏是否被切、是否过空、对比度够不够、色板有没有绕过主题变量,而不是靠肉眼翻页目测。触发词:讲稿排版检查、幻灯片验收、内容溢出、投影看不全、slide 被切、演讲页视觉检查、talk 排版、每页会不会超屏、PPT 排版验收。与 talk-text-review(查姿态/文本)互补:那个管说什么,这个管看得见看不见。

— No reviews yet
0 installs
0 views
— view→install

Install

$ agentstack add skill-webkubor-talk-skills-talk-visual-review

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • ✓ Prompt-injection patterns
  • ✓ Secret / credential exfiltration
  • ✓ Dangerous shell & filesystem operations
  • ✓ Untrusted network calls
  • ✓ Known-malicious package signatures

What it can access

  • ✓ Network access No
  • ✓ Filesystem access No
  • ✓ Shell / process execution No
  • ✓ Environment & secrets No
  • ✓ Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-webkubor-talk-skills-talk-visual-review)

Reliability & compatibility

✓ Security review passed
0 installs to date
— no reviews yet
● today

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Talk Visual Review? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

演讲页视觉验收:量出来,不是看出来

为什么不能靠肉眼

翻 19 屏靠眼睛看,三个必漏:

  1. 屏幕不一样:你在 2560×1440 的 Mac 上看着刚好,投影是 1280×800,内容当场被切。
  2. 缩放会骗人:talk 页常有 --talk-scale 之类的等比缩放,本机缩小了看着没事,投影放大就溢出。
  3. 量太大:20 屏 × 4 种分辨率 = 80 次目测,人一定会跳着看,而出事的往往是没看的那几屏。

2026-08-28 实测翻过一次车:一屏塞了表格 + 引用 + 两段正文,标题被切掉一半、底部内容跑到屏幕外,翻页也够不着——肉眼在开发机上完全看不出来。

核心原则:分屏用估算,验收用实测,两者必须独立。 用同一个估算模型既分屏又验收,等于自己给自己打分。

主检查:内容是否被切

facet 项目直接跑现成脚本:

node scripts/check-slide-overflow.mjs dist-share//talk.html
node scripts/check-slide-overflow.mjs https://share.webkubor.online//talk --size 1920x1080

非 facet 项目,核心测量逻辑(在任意 CDP 浏览器里 evaluate 即可):

() => {
  const slides = [...document.querySelectorAll('.slide')];   // 换成目标页的屏选择器
  const saved = slides.map(s => s.className);
  const vh = window.innerHeight;
  const rows = [];
  slides.forEach((s, i) => {
    slides.forEach(x => x.classList.remove('active'));       // 逐个临时激活
    s.classList.add('active');
    const body = s.querySelector('.slide-body') || s;        // 量真正的内容容器
    const r = body.getBoundingClientRect();
    rows.push({
      i: i + 1,
      fill: Math.round(body.scrollHeight / vh * 100),
      cutTop: r.top  vh + 1
    });
  });
  slides.forEach((x, i) => x.className = saved[i]);
  return rows;
}

两个坑:

  • 必须逐个临时激活。靠 .active 控制显示的页,不激活的屏 getBoundingClientRect 全是 0,量出来一片正常。
  • 量内容容器,不量 wrapper。flex 布局的外层会被拉伸到视口高度,scrollHeight 对所有屏返回同一个数——看到一列一模一样的高度就说明量错了元素。

判据

| 填充率 | 判定 | 处理 | |---|---|---| | 越界(topvh) | 🔴 被切 | 必修。讲的时候内容在屏幕外,翻页够不着 | | > 90% | 🟡 危险 | 换个分辨率就可能切,建议拆 | | 35% – 90% | ✅ 正常 | | | s.textContent) .join('').match(/#[0-9a-f]{3,8}(?![^{]*\})/gi)

// 2. 正文对比度:投影环境比显示器暗,对比度不足后排看不清 getComputedStyle(document.querySelector('.slide-body p')).color

// 3. 代码块字号:正文与代码字号差距过大是投影上最常见的「看不清」来源 getComputedStyle(document.querySelector('.slide pre')).fontSize


经验阈值:投影场景正文 ≥ 18px、代码 ≥ 16px,正文与代码字号差 ≤ 4px。对比度按 WCAG AA(正文 4.5:1),投影建议做到 7:1。

## 机器测不了的,交给人

这套只回答「**看不看得见**」,回答不了「**讲不讲得顺**」。以下必须人自己翻一遍:

- 断句是否自然(拆屏可能把一个论证从中间劈开)
- 每屏的信息重心对不对(一屏该有一个主张,不是一堆并列)
- 表格/代码在这一屏出现的时机对不对
- 续屏标记是否让人误以为翻重复了

**报告时明确分开说**:哪些是实测结论(有数字),哪些需要人确认。别把「我测了没被切」说成「排版没问题」。

## 边界

- 只验渲染结果,不改内容、不调分屏策略(那是构建器的事)
- 不替代真机彩排:投影的色温、亮度、可视角度测不出来
- 报告里给出实测数字(几屏、填充率、什么分辨率),不写「看起来还行」

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [webkubor](https://github.com/webkubor)
- **Source:** [webkubor/talk-skills](https://github.com/webkubor/talk-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.