Update README.md

README.md

@@ -13,7 +13,7 @@ tags:
 - multi-modal
 ---
 
-<h1 align="center">WildClawBench Dataset</h1>
+
 <p align="center">
   <img src="https://huggingface.co/datasets/internlm/WildClawBench/resolve/main/assets/lobster_battle.png" alt="WildClawBench Lobster" width="480">
 </p>
@@ -21,6 +21,9 @@ tags:
   <b>Hard, practical, end-to-end evaluation for AI agents — in the wild.</b>
 </p>
 
+🏠<a href="">Github</a> | 🥇<a href="">Leaderboard</a> |📝<a href="">Blog</a> | 🤗<a href="">Image/Data preparation</a> 
+
+
 ## 📌 Overview
 
 **WildClawBench** is an agent benchmark designed to test real-world utility: can an AI agent perform complex work end-to-end without hand-holding? Agents are deployed into a live [OpenClaw](https://github.com/openclaw/openclaw) environment—a real-world personal AI assistant—to tackle **60 original tasks**.
@@ -32,7 +35,7 @@ These tasks are designed to be significantly more difficult than existing benchm
 This Hugging Face repository hosts the heavy assets required to run the benchmark:
 
 * **`Images/wildclawbench-ubuntu_v1.2.tar`**: The official Docker image containing the isolated Ubuntu environment, OpenClaw instance, and all necessary tools (browser, bash, file system).
-* **`workspace/`**: The task data directory containing initial files, evaluation files, and environment configurations for all 60 tasks.
+* **`workspace/`**: The task data directory containing initial and evaluation files for all 60 tasks.
 
 ## 📊 Benchmark Structure