Prism-privacy_policy

PRISM — Privacy Policy

Applies to: PRISM — contextual gesture shortcuts (demonstrator), extension version 1.1.1 Last updated: 17 November 2026 Effective date: 2026-11-17


0. What this extension is, before anything about data

PRISM is a research demonstrator written for a Human–Computer Interaction course project at Stony Brook University. It exists to let a person try one interaction technique — direction gestures drawn with the right mouse button, with an on-screen guide that shows what the gesture would do — on real web pages instead of only inside a lab instrument.

It is not a supported consumer product. It has no company behind it, no support desk, and no guarantee of continued maintenance. No study results are claimed here, because the study has not been run yet.

This policy describes what the published extension actually does. Every statement below was checked against the extension’s source code rather than written from memory.


1. The short version


2. What is stored on your machine

   
Where chrome.storage.local — local to this browser profile on this computer
Keys exactly two: prism.custom-bindings and prism.settings
What is in prism.custom-bindings the custom gestures you recorded on the extension’s options page
What is in prism.settings three settings you choose on the options page: whether trackpad mode is on, the sensitivity (standard or larger movements), and the list of website names (e.g. example.com) where you turned PRISM off
Per binding the context (page, link, image or selection), the direction tokens (e.g. U,L), the id and label of the command it runs, and the time it was created
Written by the options page, when you save a gesture
Read by the content script, once per page load, so a saved gesture works after a browser restart

No page content, no page addresses, no page titles, no timing data and no identifiers are stored. The one free-text field on the options page is the list of sites where PRISM should stay off; it accepts website names only, and it holds exactly what you type into it. PRISM never adds a site to it by itself and never records which sites you visit.

chrome.storage.sync is not used, so nothing is uploaded to your Google account. The extension also uses no localStorage, no sessionStorage, no IndexedDB and no cookies.

Deleting a binding or a site on the options page removes it. Uninstalling the extension removes both keys and everything in them.


3. The permissions, and exactly what each is spent on

bookmarks

Spent by one call in the whole extension: chrome.bookmarks.create({ title, url }) in background.js, in the handler for the bookmark action. It is reachable only one way — by drawing the R gesture on a link, the “Bookmark link” command. The title and URL passed to it are the title and URL of that link.

The extension never reads, searches, lists, edits, moves or deletes bookmarks. create is the only bookmarks API it calls, and without this permission that one command cannot work.

storage

Spent by get and set calls on chrome.storage.local against the two keys described in §2, in bindings.js (your gestures) and settings.js (your settings), plus listening for changes so a new setting applies without reloading the page. A custom gesture has to survive a browser restart; a content script’s localStorage belongs to the website you are visiting, so it is per-site, cleared with site data, and invisible to the options page. chrome.storage.local is the only store the options page and the content script can share.

downloads

Spent by one command: Save image, the D gesture on an image. background.js calls chrome.downloads.download({ url, filename, saveAs: false }) with the address of the image you gestured on (or, for an inline SVG figure, a data: address built from its own markup, which involves no request), so the file is saved straight to your browser’s default Downloads folder without a Save-As dialog. The file name is made from the image’s description or address and cleaned of anything that is not safe in a file name. The browser makes the download request, exactly as it would for “Save image as…”; the file appears in Chrome’s own downloads list, and you can delete it like any other download.

To report truthfully whether the file actually finished, the worker also looks up that one download’s state (chrome.downloads.search by its id, and onChanged filtered to that id). It never lists, reads, opens, deletes or otherwise touches any other entry in your download history.

sessions

Spent by one call: chrome.sessions.restore(), with no argument, in background.js. It is reachable only by drawing D, L on a page, the “Reopen closed tab” command, and it reopens the tab or window you closed most recently, exactly like Chrome’s own Ctrl+Shift+T. The extension never lists, reads or searches your recently closed tabs or the tabs on your other devices.

scripting

Spent by one call: chrome.scripting.executeScript with the extension’s own content script file (extension/content.js, the same file the manifest already runs on every web page). It runs only when the extension is installed or updated, so that tabs you already had open get the gestures without being reloaded. To find those tabs, the worker asks Chrome once for the open http:// and https:// tabs; it uses the answer only to address the injection and keeps nothing. It never injects any other code, never passes it page data, and never reads anything back.

Host access: http://*/* and https://*/*

Spent by one command: Copy image, the L gesture on an image (§5). A web page is not allowed to read the pixels of a picture that came from another website, so the content script cannot turn such a picture into clipboard data itself. The extension’s background worker can, but only for websites it holds host permission for — and the picture can be on any website, which is why the grant is broad. It is used for exactly one request: a fetch of the address of the picture you gestured on, inside the Copy image handler in background.js, described in §5. The host permission is not used to read, change or inject into any page; the content script’s page access is described separately in §4.

Nothing else is requested

There is no tabs, cookies, history, identity, management, webRequest, clipboardRead or clipboardWrite permission.

The extension does open, close and create tabs and windows (including an incognito window, when you draw “New incognito window” or “Open in incognito window”), and it moves a tab back or forward in its history when you gesture inside an embedded frame. Those APIs are ungated and buy no access to your data: the tab it closes is the one the gesture came from, identified by the message sender, and when it reads the geometry of your current window in order to size a new one, it does so without populate, so no tab list, URL or title is returned to it.


4. Why the content script runs on every http:// and https:// page

The manifest declares a content script matching http://*/* and https://*/*. The technique is a replacement for the right-click context menu, so it has to be available wherever you right-drag; the extension cannot know in advance which pages those will be, and a narrower match would make it work on some sites and silently do nothing on others. It does not run on chrome:// pages, the Chrome Web Store or the new tab page, and it checks the page’s scheme itself before starting.

What it does on a page you visit:

It also runs inside embedded frames on a page (for example an embedded video or editor), so that gestures work there too; it waits until a frame is at least 200 × 150 pixels, so small advertising frames are left alone. In every frame it does exactly what is listed above and nothing more.

What it does not do: it does not read or index page text beyond the selection you have made yourself; it does not read form fields (beyond the selected text described above), passwords, cookies, localStorage or site data; it does not scrape, copy or store page content; it does not record your browsing history; it does not capture keystrokes; it does not modify the page apart from adding its own overlay; and it writes nothing about any page you visit to disk. Nothing it reads is retained after the gesture ends.

While a gesture is in progress, it also makes the page’s embedded frames (ads, videos, embeds) briefly ignore the mouse, so a gesture that passes over one isn’t lost; each frame’s original setting is restored exactly the moment the gesture ends (and after 10 seconds at most, whatever happens).

Beyond that, it changes the page’s normal behaviour in exactly two situations, both only during a gesture you are drawing. (1) After a right-drag that moved more than 10 pixels, it stops the right-click menu that would otherwise pop up at the end of the gesture; a short right-click (10 pixels or less) still opens the website’s own menu exactly as that site intended. (2) In trackpad mode, during an Alt-drag that moved more than 10 pixels, it stops text selection, link dragging and the click at the end, so the gesture doesn’t also follow the link; a short Alt+click works as normal.


5. The clipboard

Three commands write to your system clipboard, each only when you draw its gesture:

Gesture Command What is copied
L on a text selection Copy the text you selected
L on a link Copy link that link’s URL
L on an image Copy image the picture itself, as a PNG image; if that fails, the image’s address (or, for an inline SVG, its markup) instead, and the on-screen message says which

Copy and Copy link use the browser’s Clipboard API (navigator.clipboard.writeText), with the older document.execCommand('copy') as a fallback if the first is refused. Copy image uses navigator.clipboard.write with an image/png item.

How Copy image gets the picture, including its one network request:

The extension never reads the clipboard. It holds no clipboardRead permission and writes only what the command you ran is defined to write.


6. The two commands that open a third-party website

Two commands in the selection context open a new tab at a public website, with the text you selected placed in the URL:

Gesture Command Opens
D on a text selection Define https://en.wiktionary.org/wiki/Special:Search?search=<your selection>
U on a text selection Search this https://duckduckgo.com/?q=<your selection>

The selection is URL-encoded and trimmed to 200 characters. The resulting request is made by your browser to that website, exactly as it would be if you typed the address yourself or used Chrome’s own “Search with…” menu item. Those sites’ own privacy policies then apply to that visit. PRISM has no API key, no account and no relationship with either site, and it never sees or reads what comes back.

These two gestures and Copy image’s re-fetch of the picture you gestured on (§5) are the only times a request goes out because of PRISM, and each happens only when you deliberately draw that gesture. Save image’s download (§3) is likewise a request for a file you chose, made by the browser as any download is.

Likewise, the options page has one ordinary link, Rate or review PRISM on the Chrome Web Store, which opens PRISM’s own page on the Chrome Web Store (https://chromewebstore.google.com/detail/dffefhigobgkdlfhgcogefngbkffkphn/reviews) in a new tab only when you click it; the extension sends nothing and records nothing about it.


7. What PRISM never does


8. How you can check these claims


9. Retention, and getting rid of it

There is nothing to retain, because nothing is sent to PRISM and there is no server. A picture you copy stays on your clipboard until you copy something else; an image you save stays in your Downloads folder until you delete it. Your custom bindings stay on your computer until you delete them on the options page or uninstall the extension, which removes them with it. There is no account to close and no deletion request to file — but if you want to ask about any of this, the contact is in §11.

10. Children, sensitive data, and changes to this policy

The extension is not directed at children and collects nothing from anyone, so it holds no sensitive category of data — no health, financial, location, authentication or personal-communications data.

If this policy changes, the updated version will be published at the same URL with a new date at the top. An extension that started collecting data would be a materially different one, and this page would say so plainly before it shipped.

11. Contact

Questions, corrections, or anything in this policy that does not match what you observe:

Manush Patel
Department of Computer Science, Stony Brook University
manushsachin.patel@stonybrook.edu