fix: detect LinkedIn 2026 feed DOM promoted posts (v1.3.5)

[2gelbuy]

Why

Second uninstall feedback (2026-06-12): promoted posts not hidden. LinkedIn's 2026 feed rewrite (article[data-id="main-feed-card"], actor line in p[componentkey]) was invisible to the v1.3.4 detector — it skipped all <p> content as body copy and covered only ~15 of 28 locales.

What

• New-DOM containers (main-feed-card, li.feed-item) + LinkedIn's own structural sponsored markers (data-sponsored-tracking-url, quoted-token root-only data-view-tracking-scope)
• Strict-equality matching for <p> metadata; loose word-boundary matching stays limited to the shipped v1.3.4 keyword list (avoids hiding organic posts with promoted-like headlines)
• Keywords expanded to ~40 languages; 2-char CJK matched only inside LinkedIn-generated sub-description lines
• Hiding via injected CSS rule on marker attribute (+static pre-hide on LinkedIn attrs) — survives re-renders, no ad flash
• Leading+trailing 250 ms throttle replaces starvable resetting debounce
• Fixes: container-div label false positive, poll innerHTML false positive, never-matching RU newsletter keywords

Verification

• npm test 33/33, npm run typecheck, npm run build, npm run zip (manifest 1.3.5)
• Consilium council review passed (run 20260612T092359); 4 blockers fixed
• Already shipped: CWS PENDING_REVIEW 1.3.5, AMO published (0 errors), Edge submitted

🤖 Generated with Claude Code

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: bd63b042-1b39-4a2f-ab73-1278bb85e5bc

📥 Commits

Reviewing files that changed from the base of the PR and between 97f0413 and 0227e2f.

⛔ Files ignored due to path filters (1)

• package-lock.json is excluded by !**/package-lock.json

📒 Files selected for processing (7)

• AI_STATUS.md
• CLAUDE.md
• package.json
• src/entrypoints/linkedin-feed.content.ts
• src/lib/linkedin-detector.test.ts
• src/lib/linkedin-detector.ts
• wxt.config.ts

---

📝 Walkthrough

Summary by CodeRabbit

• New Features

  • Enhanced detection of promoted and sponsored content across LinkedIn feeds.
  • Improved handling of promoted labels in multiple languages and formats.
  • Support for newer feed DOM structures.

• Bug Fixes

  • Improved reliability of hiding promoted content during feed re-renders using stylesheet-based approach.

• Tests

  • Expanded test coverage for feed post detection and promoted label hiding scenarios.

Walkthrough

Extension version 1.3.5 promotes promoted-post hotfix improvements. The detector gains expanded DOM selectors, stronger text normalization, and multi-tier keyword matching. The content script switches to stylesheet-based element hiding with refined mutation scheduling and selective observer filtering.

Changes

LinkedIn Feed Filter Enhancement v1.3.5

Layer / File(s)	Summary
Version and metadata updates wxt.config.ts, package.json, AI_STATUS.md, CLAUDE.md	Extension version updated from 1.3.3 to 1.3.5. Release status and local documentation paths adjusted for v1.3.5 hotfix cycle tracking.
LinkedIn detector infrastructure and text normalization src/lib/linkedin-detector.ts (selectors, normalization, keywords)	DOM selectors expanded for 2026 LinkedIn patterns. Text normalization strengthened with NFKC, case-folding, Unicode cleanup, and broader punctuation stripping. Keyword lists split into loose/strict matching tiers for promoted/suggested/newsletter detection.
Label collection and keyword matching src/lib/linkedin-detector.ts (label pipeline)	Label candidates collected from DOM elements and aria-describedby references with exclusion rules. Loose keyword arrays normalized and strict Sets generated. Metadata-line token matching implemented with max-length constraints.
Promoted and suggested signal aggregation src/lib/linkedin-detector.ts (signal functions)	Strict metadata extraction via tokenized matching. hasPromotedLabel and hasSuggestedLabel combine strict/loose label checks with metadata and metadata-token matching.
Feed discovery and hiding logic src/lib/linkedin-detector.ts (integration)	Feed post detection incorporates structural markers and label signals. Poll detection simplified to DOM-role/class checks. shouldHideLinkedInPost checks both structural markers and label signals.
LinkedIn detector test expansion src/lib/linkedin-detector.test.ts	New tests cover feed containers, newer URN families, promoted label paths (actor divs, metadata, aria-describedby), organic exclusions, main-feed-card DOM, CJK language/format cases, poll handling with hidePolls flag, and interaction-scope regressions.
Content script stylesheet and element management src/entrypoints/linkedin-feed.content.ts (stylesheet, restore)	Stylesheet-based hiding system with HIDE_STYLE_ID. Element restore refactored to use marker attributes and PREVIOUS_DISPLAY_ATTR instead of inline snapshots.
Content script processing pipeline and scheduling src/entrypoints/linkedin-feed.content.ts (processing, throttle)	Unified runProcessPass() applies stylesheet, processes posts, and hides sidebar ads. Scheduling switches from fixed delay to leading+trailing throttle (immediate if >600ms, else 250ms wait). Storage listener triggers filter updates.
Content script mutation observer refinement src/entrypoints/linkedin-feed.content.ts (observer)	Selective filtering via shouldProcessMutation() ignores irrelevant mutations. Observer expands to include characterData and filtered attributes. Initial run switches to runProcessPass().

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/linkedin-2026-dom-v135

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch fix/linkedin-2026-dom-v135

• 🛠️ tighten validation

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}