Dark Pattern Detector

Privacy-focused Chrome extension that detects and highlights manipulative interface patterns while you browse.

Current implementation

The extension is a Manifest V3 prototype built around a local-first detection pipeline:

• Regex matching against configurable phrases in patterns.txt
• 13 text categories, including urgency, scarcity, social proof, hidden costs, forced action, trick questions, click bait, and curiosity gaps
• Optional visual-interference detection for weakly presented opposing actions such as a faint “Reject” button beside a prominent “Accept” button
• Optional on-device Gemini Nano verification through Chrome’s Prompt API
• Automatic scanning on page load and incremental scanning of dynamically added page content
• Popup dashboard with counts, category details, pause/resume, and detection-mode status

When Gemini Nano is unavailable, the extension continues to work in regex-only mode. Website text is not sent to a project-operated cloud service.

Install locally

1. Clone the repository:

   git clone https://github.com/GamePointAnalytics/dark-pattern-detector.git

2. Open chrome://extensions in Chrome.

3. Enable Developer mode.

4. Select Load unpacked.

5. Choose the dark-pattern-detector-chrome-extension directory.

6. Refresh any already-open tabs you want to scan.

The extension works without Gemini Nano. Experimental Chrome AI features must be enabled separately if you want optional AI verification; see the extension README for current setup notes.

Project structure

dark-pattern-detector-chrome-extension/
├── content.js DOM scanning, highlighting, heuristics, and AI requests
├── patterns.txt Detection categories and regex fragments
├── popup.html/js Extension dashboard and controls
├── background.js Service-worker router and offscreen lifecycle
├── offscreen.js Optional Gemini Nano integration
├── styles.css Highlight styles
├── test_*.html Manual test fixtures
└── report/ Design and implementation notes

Development and testing

There is currently no build step. The pure detector and popup feedback paths have lightweight Node regression tests. For basic validation:

• Run node --check on the JavaScript files after edits.
• Run node dark-pattern-detector-chrome-extension/test_detector_core.js to verify the pure text-detection module.
• Run node dark-pattern-detector-chrome-extension/test_popup_feedback_persistence.js to verify feedback survives popup reopening.
• Run node dark-pattern-detector-chrome-extension/test_active_tab_capture.js to verify the consent-gated, one-shot active-tab capture boundary without retaining an image.
• Run node dark-pattern-detector-chrome-extension/test_image_signal_core.js to verify that only approved image-signal categories and confidence values can be retained.
• Run node dark-pattern-detector-chrome-extension/evaluate_detector.js to print category-level development-fixture precision and recall.
• Load the extension unpacked in Chrome.
• Use test_page.html and test_context.html to exercise regex detection and context handling.
• Use test_active_tab.html to exercise the explicit active-tab screenshot and on-device image-analysis flow with safe visual examples.
• Use test_nano.html only when testing Chrome’s optional built-in AI capability.

Detection patterns can be edited directly in dark-pattern-detector-chrome-extension/patterns.txt; reload the extension and refresh the test page after changing them.

Current limitations

• English-language patterns only
• Regex matching can produce false positives for benign text
• Visual analysis currently focuses on selected opposing-action button pairs
• Rapidly changing single-page applications may not be scanned immediately
• Gemini Nano availability depends on the Chrome version, device, flags, and model state
• This is a prototype and should not be treated as a complete accessibility, safety, or legal compliance tool

Documentation

• Product direction research
• Capture modes and consent spec
• Legal and research readiness
• Product frames
• Threat model
• Data inventory
• Local event schema
• Product consent and local data view
• Active-tab analysis beta
• Secure AI analysis options
• Classifier evaluation plan
• MVP quality checklist
• Product roadmap
• Research backlog
• Domain context
• Architecture decision: local-first analysis
• Architecture decision: user-controlled capture modes
• Extension documentation
• Walkthrough
• Implementation report
• Resources and references

License

The repository currently does not include a license file. Licensing should be added before distributing the project publicly.