Phase 3: polish

[rcholic]

7. Implementation Roadmap

Approach: Ship v0 minimal slice first (see Appendix B.10), then expand.

Phase 1: v0 Proof of Concept

• Implement CDPTransport protocol and BrowserUseCDPTransport
• Implement CDPBackendV0 (10 methods from B.10.1)
• Implement BrowserUseAdapter.get_transport()
• Load Sentience extension into browser-use via --load-extension
• Upgrade click_async with scroll handling (B.10.7)
• Test on 3-5 representative pages

Phase 2: Full Integration

• Add type_async, scroll_async actions using Sentience grounding
• Refactor existing actions.py and snapshot.py to use backend protocol
• Create PlaywrightBackend wrapper for backwards compatibility
• Add snapshot caching with staleness detection
• Write comprehensive tests

Phase 3: Polish

• Documentation and examples
• Performance benchmarking vs. baseline browser-use
• Error handling for extension loading failures

---

8. Risks and Mitigations

Risk	Likelihood	Mitigation
Extension injection fails (CSP)	Medium	Fall back to CDP-only mode
Snapshot too slow (<500ms target)	Low	Cache snapshots, only refresh on navigation
API key management complexity	Low	Use env vars, match browser-use patterns
CDP vs Playwright API drift	Low	Abstract via adapter layer