The story is a first-person account, not a product audit. It is still a useful case study in AI-assisted development. It also raises a question every user of a cross-tab tool should ask: what does this extension need to see in order to work?
The problem: Ctrl+F has a one-tab horizon
Ctrl+F (Cmd+F on a Mac) searches only the page you are looking at. Faisal describes the familiar research-day scenario. You remember a statistic or a quote, and you have dozens of tabs open. You can't recall which tab holds it, so you click through them one at a time.
She says she looked for an existing extension that fit her workflow and didn't find one. She also acknowledges that one may exist. This is not a claim that nobody has solved the problem. It is an account of one person who decided to build her own fix.
What she asked for, and what she says she got
Faisal says she used Claude Code on the web in Auto mode with the Fable 5.2 model. She gave it one detailed prompt. The prompt asked for:
- A Spotlight-style search box that appears over the page you are viewing.
- Search of the actual page text, not just tab titles or URLs.
- Snippets of matching text, with the match highlighted.
- A jump to the exact spot on the page where a match sits.
- Filters by website, tab group, or browser window.
- Exact-phrase search, case-sensitive search, and search history.
- Fully local operation, with no API key, backend, or AI model.
Claude Code reportedly built the search engine first, then the interface, icons, and tests. Along the way it hit failing browser tests and a bug that wiped text typed into one of the filters. According to the article, it found and fixed both without her stepping in.
The reported test and performance numbers
The article says Claude ran 25 browser checks and 13 unit tests, and all passed. It also reports a stress test with 40 tabs of 250,000 characters each, scanned in roughly 0.7 seconds.
Treat these as the author's account of what the AI reported about its own project. They are not independently reproduced benchmarks. They do show that the tool tested itself, which is one real advantage of agentic coding tools over a plain chatbot.
Day-to-day use
Faisal says she has used the extension for several days, often with more than 40 tabs open. In one test she searched a 271-page Google document and found text from the middle of it within seconds. In another, a search for "prompt" across 19 tabs returned 21 matches in five tabs.
Results are grouped by tab with a count for each. You can expand a tab to see each occurrence with surrounding text. Clicking a match takes you to that spot on the page. The extension can search the current window or all Chrome windows, and you can hide chosen sites from results.
She says the extension skips Chrome's internal pages, such as the Extensions page and the New Tab page. She also says she uploaded the code to a GitHub repository.
The part worth slowing down on: permissions
An extension that reads the text of every open page is powerful. That power is also why you should look at it carefully. The Chrome documentation explains how this access works.
- Host permissions control page access. Chrome's developer documentation says an extension needs host permissions for a page before it can inject a content script into it. Those permissions can be declared in the manifest, or granted temporarily through "activeTab".
- activeTab is narrow by design. The "activeTab" permission gives an extension temporary access to the currently active tab when the user invokes the extension. Chrome's tabs documentation adds that unlike host permissions, activeTab does not trigger any warnings. That makes it a poor fit for a tool that has to search every tab at once, which probably needs broader access.
- Broad access can come with a standing grant. A Chrome Dev Journey audit guide notes that all-site grants show the warning "Read and change all your data on all websites". It describes that access as standing once granted.
- Users can narrow access after install. Chrome 70 let people restrict an extension's host access to a custom list of sites, according to PPC Land's summary of the permission model.
None of this says Find Everywhere is unsafe. The XDA article doesn't describe its manifest, its permission prompts, or what it stores. "Local" tells you where the search runs. It does not tell you how much the extension can see, or whether it keeps a record of your searches.
Why internal pages are off limits
The limitation Faisal describes matches Chrome's documented behavior. With activeTab, access is not granted to restricted pages, such as chrome:// pages. The Chromium extensions security FAQ also says chrome:-scheme pages such as chrome://settings are generally protected from extensions. Extensions can open and close them, but cannot read or change their data.
An extension-permissions explainer from extpower.dev goes further. It says neither activeTab nor broad host permissions grants injection into Chrome internal pages or the Chrome Web Store. A cross-tab search tool can therefore never be fully universal, however it is built. The same source adds that an enterprise policy can also block access. IT admins who manage Chrome through policy should expect gaps on locked-down machines.
A practical checklist before installing any cross-tab extension
If you try Find Everywhere, or any extension that scans many tabs, a few minutes of checking is worthwhile:
- Open the extension's manifest and read the
permissions,host_permissions,optional_host_permissions, andcontent_scriptsmatchesfields. These show what the extension can actually reach. - Check whether it uses
<all_urls>or similar broad patterns. If it does, find out why. - Check whether it requests the
storagepermission, and look for where search history lives. Local storage is common, but you should know what is kept. - After installing, use Chrome's site-access controls to limit it to specific sites if you don't need it everywhere.
- Try it on a chrome:// page and on a page you don't care about, to see where it stops.
- Skim the source for any network calls, because "no backend" is a claim you can check in code.
For code you didn't write, this is also a fair general rule. AI-generated code is still code. If you load an unpacked extension from a repository, read it first.
What this says about AI-assisted development
This story is a good example of the sort of project that AI coding tools suit well. It is small and self-contained, and the author could check the result against her own workflow. She was also clear that it isn't complicated or groundbreaking.
Faisal says the model name matches Anthropic's Fable lineup. Anthropic's product page describes Fable 5.2 as its most capable model for technical work. The page says it is available to Pro, Max, Ultra, Team, and Enterprise users, and that fast mode is also available on Claude Code. Those details only confirm the model exists. They don't confirm how Faisal used it.
There are fair caveats:
- It's one person's workflow. A tool that fits a journalist with 40 research tabs may not fit everyone.
- The self-testing is self-reported. Passing tests written by the same AI that wrote the code is encouraging. It isn't a security review.
- Maintenance is unaddressed. Browsers change extension rules over time, and a one-evening build still needs someone to keep it working.
The bottom line
Cross-tab text search is technically plausible with standard extension tools, and Chrome's documented model explains both what it can do and where it must stop. The XDA piece shows how quickly an idea like this can become a working prototype. It does not show that the result is audited, benchmarked, or unique.
If you build or install something like this, treat the permission list as the real feature list. Whether a tool searches all your tabs matters less than what else it can read while it does.
References
- Ctrl+F only searches one tab, so I used Claude Code to build an extension that searches them all XDA · 2026-11-19T11:00:21+00:00
- Claude Fable \ Anthropic anthropic.com
- The "activeTab" permission | Chrome Extensions | Chrome for Developers developer.chrome.com