Add ClawBench evaluation framework

[reacher-z]

Summary

• register clawbench-eval as a supported benchmark evaluation framework
• link the canonical ClawBench repository

The ClawBench benchmark dataset already declares this identifier in its root eval.yaml; adding it here satisfies the documented framework-enum requirement for native Hub benchmark registration.

Validation

• pnpm --filter @huggingface/tasks format:check
• pnpm --filter @huggingface/tasks check

Disclosure

I help maintain ClawBench. This focused contribution was prepared with assistance from OpenAI Codex.

---

Note

Low Risk
Single registry entry with no runtime or security impact; only extends the allowed framework list for eval.yaml.

Overview
Registers clawbench-eval in the EVALUATION_FRAMEWORKS registry in eval.ts, alongside existing frameworks like claw-eval and researchclawbench. The entry includes a short description of ClawBench (web-agent evaluation with isolated runs and behavioral traces) and links to https://github.com/TIGER-AI-Lab/ClawBench.

This aligns the tasks package with benchmark datasets whose root eval.yaml already reference this framework identifier for native Hub registration.

^{Reviewed by Cursor Bugbot for commit 87e3439. Bugbot is set up for automated code reviews on this repo. Configure here.}