FeedKit

What I personally did

• –Product definition, store listing, and privacy policy.
• –Manifest V3 extension: content script, background worker, popup, and options page.
• –Post detection that survives Facebook's generated class names, plus the keyword, page, and rule engine.
• –Video controls, reading and zen modes, scroll and daily time limits, and on-device usage stats.

• –Facebook's markup uses generated class names that change between deploys, so nothing can be selected the obvious way.
• –The feed is infinite and rendered on the fly, so filtering has to keep up with posts as they appear without making scrolling stutter.
• –Privacy was a product requirement: no server, no account, and nothing about the user's feed may leave the browser.
• –Manifest V3 rules: no remote code and a background worker that can be shut down at any time.

1. Detect by meaning, not by class name

   Facebook's class names are generated and change without notice, which breaks most feed-cleaning extensions within weeks.

   Considered

   Target the current class names and patch them whenever they change.

   Chose

   Identify posts and surfaces from ARIA labels, roles, data-pagelet markers, URL patterns, and visible text, the signals the page needs for accessibility and routing.

   Trade-off

   More heuristics to write and test than one selector, but a Facebook deploy rarely breaks the extension.

2. Two mechanisms: CSS for surfaces, a scanner for posts

   Some things never change position (the Reels tray, sidebars); posts stream in endlessly.

   Considered

   Run every rule through one JavaScript scanner.

   Chose

   Static surfaces are hidden with injected CSS before first paint. Only posts go through the MutationObserver and the rule engine.

   Trade-off

   Two code paths to maintain, but nothing flickers on load and the scanner only does work that needs judgement.

3. No server at all

   An extension that reads your feed is exactly the kind of thing people should be suspicious of.

   Considered

   A small backend for accounts, stats, and remote rule updates.

   Chose

   Everything runs locally. Settings sync through the browser's own storage, stats never leave the device, and the optional AI features call the user's chosen provider directly with their own key.

   Trade-off

   No usage analytics to learn from, so feedback comes from store reviews. In return, the privacy policy is short and easy to verify.

• 92

  Users on the Chrome Web Store

  From the store listing on 19 Oct 2026.

• 4.3

  Average rating out of 5

  Six ratings on the Chrome Web Store, 19 Oct 2026.

• The whole extension ships in 108 KiB with no remote code, as Manifest V3 requires.