Support tool-based benchmark evaluation

[RuishanFang]

Description:

To support a variety of benchmark tests, our framework needs to support evaluations on more benchmarks, especially those that require tool-based evaluation (e.g., swebench, webarena).

Proposed Benchmarks to Support:

• GAIA
• SWE-bench
• WebArena
• HotPotQA

Considerations:

• The performance of different agents can vary across benchmarks. Developers should select appropriate benchmarks based on the characteristics of the agent system.