Add metadata and links to paper and code

This PR adds the missing `task_categories` metadata (`video-text-to-text`) and adds links to the paper and the official code repository to the dataset card.

README.md

@@ -1,5 +1,12 @@
+---
+task_categories:
+- video-text-to-text
+---
+
 # VideoGAIA-271
 
+[Paper](https://huggingface.co/papers/2608.14718) | [Code](https://github.com/zfkarl/VideoGAIA)
+
 VideoGAIA-271 is a human-verified benchmark for agentic video understanding. It contains 271 open-ended tasks that require a model to combine visual-temporal evidence with targeted tool use and external knowledge.
 
 ## Files
@@ -53,7 +60,6 @@ Example:
 {"sample_id":"vdr_example","video":{"local_path":"videos/vdr_example.mp4","duration_sec":21.2},"question_en":"...","answer":"...","category":"Technology","difficulty":"L1"}
 ```
 
-
 ## Loading
 
 ```python
@@ -72,4 +78,4 @@ The companion `videogaia` repository contains the official agent-loop inference
 
 ## Citation
 
-Citation metadata and the final redistribution terms should be added alongside the public VideoGAIA paper release.
+Citation metadata and the final redistribution terms should be added alongside the public VideoGAIA paper release.