How does it handle context bloat between the browser and the llm?
Any plans of exposing more of the browser? For instance playwright is able to store tracing files the agent may decide to read to understand some requests / payloads…
Any plans on allowing the agent to run an arbitrary js script?