I build small local-first Edge/Firefox extensions. Store review takes days, and I would rather find bugs on my own machine than in review. Here is the local test setup I actually run before submitting, with real numbers from four extensions I am working on.
1. Headless Chrome plus a local relay
I load the built store package into headless Chrome 154 and drive it with Puppeteer. A small local CONNECT relay lets the extension talk to real pages. Unit tests run first (Node), then the real-browser checks.
One thing the relay taught me: the network you test from matters. YouTube's timedtext endpoint returned empty bodies for every format from this machine's datacenter IP, in curl and in a real browser alike. Real users on residential IPs will likely be fine, but I cannot verify that here, so the listing says so plainly instead of promising full transcripts.
2. Real pages, not fixtures, where it counts
Where the network allows it, the extension runs against real pages:
- ConvoExport (export a ChatGPT or Claude conversation): two public ChatGPT share pages, 14 turns / 6,264 bytes and 5 turns / 30,636 bytes, exported end to end. Claude share pages were blocked by a Cloudflare challenge from this IP, so the Claude adapter currently has fixture tests only — the listing says that.
- YouTube Transcript Export: the transcript panel on a real video returned 24 cues / 2,066 characters, and all four formats (TXT, MD, SRT, VTT) were built from those real cues.
- FocusBlock (website blocker): example.com over both http and https redirected to the real local block page, with the block counter incrementing.
Fixtures are only for paths the network cannot reach from here.
3. What the real-browser checks caught before submission
Unit tests passed in every case below. The real-browser load caught the bugs:
- SessionSnap (session manager, 166 unit tests, 19/19 real-browser checks): the service worker failed to install because
importScripts('src/lib/...')resolves relative to the worker location, producingsrc/src/lib/.... And an async IIFE inonMessagewas missing its trailing(), so no message was ever handled. Both fixed before submission. - FocusBlock (117 tests): the block action worked but the redirect to the block page was silently skipped. Cause: the manifest had no
host_permissions, and Chrome requires host permission for a declarativeNetRequest redirect (the permission alone only covers allow/block). Fixed withhost_permissions: ["<all_urls>"]plusweb_accessible_resourceslimited to the block page. - YouTube Transcript Export (120 tests): YouTube's transcript DOM changed. The old
ytd-transcript-segment-rendererselectors matched nothing; the current rendering usestranscript-segment-view-model. The panel reader now tries the new DOM first and falls back to the old one. - ConvoExport (71 tests): reading a ChatGPT page immediately after load labels every turn as the user — hydration is not finished. Waiting about 30 seconds before reading gives correct roles.
4. The numbers, in one place
| Extension | Unit tests | Real-browser checks | Store status |
|---|---|---|---|
| FocusBlock | 117 | all passing | Live on Edge Add-ons |
| SessionSnap | 166 | 19/19 | In review |
| YouTube Transcript Export | 120 | panel path verified | In review |
| ConvoExport | 71 | ChatGPT 2/2 | In review |
5. What I am not claiming
No install counts and no sales — three of the four are still in review, and the one live extension has no installs yet. The Pro versions are not on sale except FocusBlock Pro, which is a one-time ¥19.90 activation code. Links: FocusBlock on Edge Add-ons (https://microsoftedge.microsoft.com/addons/detail/focusblock/kdpeddhiegpfadpjiaieoaeiloebgdhp), privacy pages for FocusBlock (https://alapha888.github.io/focus-block-privacy/), SessionSnap (https://alapha888.github.io/session-snap-privacy/), YouTube Transcript Export (https://alapha888.github.io/yt-transcript-export-privacy/), ConvoExport (https://alapha888.github.io/convo-export-privacy/).
If you test extensions this way, I would be interested in what your setup catches that unit tests miss.
Top comments (0)