I can't run your page, so everything below is a hypothesis. Each one comes with a check that will confirm or rule it out. Start at 375px, then test 320, 414, 768, 1024 and 1280, and 200% zoom at desktop width.
First, find the widest element. Run the overflow-finder snippet from the skill in the console at 375px. Skip anything inside the table wrapper. The URL card or the table will probably be at the top of the list.
1. Card with a long URL in a flex row (sideways scroll at 375px)
- Cause: a flex item has an automatic
min-width, so it can't shrink below its content. An unbroken URL has no break opportunity, so the card keeps the URL's full width and pushes the row past the viewport. - Fix:
If the row should stack on phones, add.card { min-width: 0; } /* let the flex item shrink */ .card .url { overflow-wrap: anywhere; } /* allow breaks in the URL */flex-wrap: wrapor switch toflex-direction: columnbelow your breakpoint. - Verify: at 375 and 320, the card's
getBoundingClientRect().rightshould be no more than the viewport width, and the URL should wrap inside the card. Test with the longest real URL you have. Removemin-width: 0once to confirm the overflow comes back. That proves this was the cause.
2. Table with 8 columns
- Cause: tables don't shrink below their content width, so 8 columns exceed 375px.
- Fix: scroll the table inside its own container, not the page.
<div class="table-scroll"><table>…</table></div>
If the table is read often on phones, restack each row into a card below about 600px. Don't hide columns. If the wrapper sits inside a flex row, it may also need.table-scroll { overflow-x: auto; max-width: 100%; }min-width: 0. - Verify: the page has no horizontal scrollbar at 375, and only the table area scrolls sideways. Check that every column can be reached and that the wrapper can be scrolled with the keyboard. Add
tabindex="0"to the wrapper if it can't. At 1280 the table should look unchanged.
3. Hero with height: 100vh too tall on phones
- Cause:
100vhis the largest viewport height, measured with the browser bars retracted. While the address bar is showing, the hero is taller than the visible area. - Fix:
Use.hero { min-height: 100vh; /* fallback */ min-height: 100svh; /* or 100dvh if you want it to follow the bars */ }min-heightinstead ofheightso tall content isn't clipped.svhalways fits with the bars showing.dvhresizes as the bars move, which can cause jank. Also check for100vwanywhere on the hero. It can cause sideways overflow of its own. - Verify: this needs a real phone, because emulation doesn't reproduce the browser bars reliably. Load the page with the address bar visible and confirm the hero fits without cutting off its content. Then scroll and confirm nothing jumps.
4. Sticky sidebar doesn't stick on desktop
- Cause: the parent has
overflow-x: hidden. Anyoverflowvalue other thanvisiblemakes that element the sticky element's scroll container. Because it never scrolls vertically, the sidebar has nothing to stick against. Less likely, but worth checking: the sidebar has notopvalue, or its parent isn't taller than the sidebar. - Fix: remove
overflow-x: hiddenfrom that parent, or move it somewhere that isn't an ancestor of the sidebar. If it was there to hide the overflow from bug 1 or 2, fixing those removes the need for it. If you must clip,overflow-x: clipdoesn't create a scroll container, so sticky keeps working..sidebar { position: sticky; top: 1rem; align-self: flex-start; }align-self: flex-startmatters if the parent is a flex or grid container. A stretched sidebar is as tall as its parent and has no room to stick. - Verify: at 1280, scroll the page and confirm the sidebar stays pinned. In DevTools, walk up the ancestor chain and check that none has an
overflowother thanvisible. Then check 375 again, since removing the clip could expose any overflow that remains.
Order of work
Fix 1 and 2 first and re-run the finder. Then fix 4, which removes the clip that was hiding the problem. Without both, you'd be hiding the overflow while breaking sticky. Do 3 last, on a real device.
I haven't tested any of this. If you can paste the computed styles for the card, the table wrapper and the sidebar's ancestors (not your code), I can narrow the causes down further.
Real run in an isolated folder with only this skill installed. Only the Skill and Read tools were enabled, so no command was run and no file was written. The scenario is fictional and the model had no access to the code. Everything in the answer is a hypothesis with a check to confirm it.
它做什么
一套排查在某些屏幕宽度下才出问题的布局的方法。它坚持先在一个明确的宽度下复现问题,再用一段控制台代码列出比视口更宽的元素,找到确切溢出的那个元素,然后才去改 CSS。一张“症状—原因”对照表涵盖了常见的元凶:固定宽度和 100vw(它包含滚动条宽度)、有自动最小尺寸的 flex 和 grid 子项(min-width: 0)、很长的不间断字符串、图片、表格、被祖先 overflow 或过矮的父元素挡住的 position: sticky、被困在层叠上下文里的 z-index、被裁掉的下拉菜单、手机上的 100vh(dvh 和 svh)、安全区(viewport-fit=cover 和 env())、grid、viewport meta 以及布局偏移。
还涵盖
在内容出问题的地方选断点、移动优先的写法、相对单位、容器查询、触控目标、200% 文字缩放和 320 像素重排,以及要测试的宽度和条件清单。它还提醒不要靠在 html 或 body 上隐藏溢出来“修复”。
适合什么场景
手机端布局 bug、“只在某个宽度才出问题”,以及会横向滚动的页面。
低风险:纯指令文件,没有脚本,不联网、不写文件。其中有一小段让你粘贴到浏览器控制台的代码,只读取元素位置,不改变页面。它看不到你的页面,所以给出的都是带有验证办法的假设,并说明没有测试的部分;模拟器无法可靠还原浏览器工具栏和安全区,涉及这些时请用真机测试。较新的 CSS 特性(容器查询、`dvh`)需要对照你的浏览器核对支持情况。AIBars 原创(MIT)。已用一个虚构页面试用过一次。