Why two browser tools? agent-browser VS openclaw-browser

[shockylove]

Description

Hi WildClawBench team,

I noticed that both agent-browser and the built-in openclaw-browser are available as skills. While they share similar capabilities, the SKILL.md suggests different use cases.

I have a few quick questions regarding their design:

• Redundancy: Why provide both? Is it to test if agents prefer AXTree/JSON (agent-browser) over Visual/Screenshot (built-in) feedback?
• Observations: During your benchmarks, did you find cases where one tool consistently solves tasks that the other cannot?
• Future: Do you plan to merge them, or keep them separate to evaluate "modality selection" abilities?

I'd love to hear your insights on the rationale behind this dual-tool setup. Thanks!