Local Gestures

English · 简体中文

Mouse gestures for Chrome that never talk to a server.

Zero network requests. No analytics. No account. No remote config. Nothing about your browsing leaves the machine — and you can verify that yourself in about ten seconds (see Verifying the privacy claim).

Manifest V3 native. English / 简体中文 / 繁體中文.

---

Why this exists

Manifest V2 stopped running in stable Chrome with version 138 (July 2025), and Google removes the last MV2 extensions from the Web Store on 31 August 2026. A lot of long-standing gesture extensions did not make the transition, and users were pushed toward whatever survived.

Gesture extensions are in an unusually sensitive position: to work at all, they need to see mouse input on every page you visit. That is exactly the access you would want an analytics pipeline to have. So the interesting question is not "which one has the most features" — it is "can I check what it does with that access?"

This one is built so that you can.

Features

Everything below is implemented and covered by the end-to-end test suite.

Trigger	What it is	Default
Stroke gestures	Hold the gesture button and draw. 19 gestures pre-bound, up to 3 strokes each	on
Rocker gestures	Hold one mouse button, click the other	on
Wheel gestures	Hold the gesture button and turn the wheel	on
Super drag	Drag a link / image / selection — the direction decides the action	on
Popup wheel	Hold still and eight actions fan out around the cursor	off
Simple drag	Straight left-button drag on empty page area	off
Double click	On empty page area only — double-clicking text still selects the word	off
Touch gestures	Multi-finger strokes, sharing the mouse gesture table	off
Context menu	Put chosen actions in the right-click menu	off
Toolbar icon	Click it to run any action	opens settings
Keyboard shortcuts	Four slots, bound in Chrome's own shortcut settings	on

45 actions across navigation, tabs, windows, scrolling, zoom and page tools.

Also: live gesture trail, action name shown while you draw, per-site disable list, settings import/export, and a drawing pad in the options page so you can record a new gesture by drawing it rather than typing DRU.

Defaults worth knowing

Gesture	Action
← / →	Back / Forward
↑ / ↓	Scroll to top / bottom
↓→	Close tab
↓↑	Reopen closed tab
↑↓	Reload
←→ / →←	Next / previous tab

Hold right and click left to go back; hold right and scroll to switch tabs.

Verifying the privacy claim

Do not take the word of a README. Check it:

# 1. No network APIs anywhere in the source
grep -rn "fetch(\|XMLHttpRequest\|WebSocket\|EventSource\|sendBeacon" src/
#    -> only the comment that promises there are none

# 2. No host permissions, no request interception, no history access
grep -n "host_permissions\|webRequest\|\"history\"\|\"cookies\"" manifest.json
#    -> no matches

# 3. Or just run the checker, which asserts all of the above
python tools/verify.py

tools/verify.py fails the build if a network call, a broad permission, an untranslated string, or a dangling action reference ever appears.

What it does ask for, and why

Permission	Why	Could it be smaller?
<all_urls> content script	A gesture must be recognisable on whatever page you are on	No — this is inherent to gesture extensions. It is also why the questions above matter
tabs	Close / switch / pin / duplicate tabs, read the title for "copy title"	No
sessions	Reopen closed tab	No
search	Drag-to-search using your own default engine — no search URL is hard-coded here	No
contextMenus	Optional right-click entries	No
downloads	Optional. Only requested if you set a drag direction to "download"	Yes — off by default
bookmarks	Optional. Only requested if you bind the "Add bookmark" action	Yes — off by default

Settings are stored with chrome.storage.local. Not sync — your gesture configuration is not uploaded to a Google account either.

Install

From source (until the Web Store listing is live):

1. Download the ZIP: < > Code → Download ZIP, then extract it somewhere permanent
2. Open chrome://extensions
3. Turn on Developer mode (top right)
4. Click Load unpacked and pick the folder containing manifest.json

👉 Step-by-step guide with screenshots-level detail → — including what to do about the "disable developer mode extensions" prompt, how to update, and how to back up your settings.

Works in any Chromium browser with MV3 support: Chrome, Edge, Brave, Vivaldi. Requires Chrome 116+.

Known limitations

Stated up front rather than discovered later:

• Gestures do not cross iframe boundaries. Start a stroke in the page and drag into an embedded frame and the stroke is cut short. This affects every gesture extension built on content scripts.
• No gestures on chrome:// pages, the Web Store, or other extensions' pages. Chrome forbids content scripts there; nothing can be done about it.
• Simple drag and double click are off by default because they compete with selecting text. Turn them on only if that trade-off suits you.
• Four directions only (no diagonals). Diagonals roughly double how precisely you have to draw, for very little gain.

Development

python tools/verify.py     # static consistency + privacy checks
python tools/e2e_test.py   # loads a real Chrome, drives every input mode
python tools/demo.py       # opens a browser on the demo page to try by hand

The e2e suite asserts observable effects — URL changed, tab count changed, scroll position reached the document's own maximum — rather than "the handler ran". 109 checks. Two actions cannot be driven by automation at all and are listed in MANUAL-CHECKS.md rather than quietly skipped.

⚠️ If every page looks structurally broken, check the browser before the code. A Playwright Chromium install that is missing resources.pak starts without the user-agent stylesheet, so every block element computes to display: inline — tables collapse to one line and <style> tags are painted as body text. Reproduces with no extension loaded at all. Detect with getComputedStyle(document.body).display (must be block); fix with python -m playwright install --force chromium, run to completion. gen_screenshots.py asserts this before capturing anything.

Adding an action means touching four places, and verify.py will tell you if you miss one:

1. src/common/actions.js — declare id, group, and where it runs
2. src/background/service-worker.js or src/content/content.js — implement it
3. tools/gen_locales.py — add the three translations, then re-run it
4. optionally a default binding in src/common/defaults.js

Translations live in one table in tools/gen_locales.py and are generated into all three _locales folders, so the key sets cannot drift apart. Icons likewise come from tools/gen_icons.py.

Feedback

Bugs and feature requests: please open an issue. If a gesture does not fire on a specific site, include the URL — that is almost always an iframe or a page that swallows mouse events, and it is worth documenting either way.

Licence

MIT — see LICENSE.