Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,34 @@
|
|
| 1 |
-
---
|
| 2 |
-
license:
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-4.0
|
| 3 |
+
|
| 4 |
+
configs:
|
| 5 |
+
- config_name: default
|
| 6 |
+
data_files:
|
| 7 |
+
- split: test
|
| 8 |
+
path: "data/test.csv"
|
| 9 |
+
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# CAP
|
| 13 |
+
|
| 14 |
+
CAP is a benchmark for evaluating browser agents on cross-site, human-like web tasks that require non-trivial UI interactions and visual perception of dynamically rendered content.
|
| 15 |
+
|
| 16 |
+
> This release contains the public subset of CAP. A held-out private subset is reserved for contamination-resistant evaluation; see the GitHub repo for details.
|
| 17 |
+
|
| 18 |
+
## 🔗 Links
|
| 19 |
+
|
| 20 |
+
- [🏠 Homepage](https://your-org.github.io/CAP)
|
| 21 |
+
- [🏆 Leaderboard](https://your-org.github.io/CAP/#leaderboard)
|
| 22 |
+
- [📖 Paper](https://arxiv.org/abs/XXXX.XXXXX)
|
| 23 |
+
- [💻 Code](https://github.com/your-org/CAP)
|
| 24 |
+
|
| 25 |
+
## 📝 Citation
|
| 26 |
+
|
| 27 |
+
```
|
| 28 |
+
@misc{cap2026,
|
| 29 |
+
title = {CAP: A Cross-site Browser Agent Benchmark for Complex Actions and Perception},
|
| 30 |
+
author = {...},
|
| 31 |
+
year = {2026},
|
| 32 |
+
eprint = {arXiv:XXXX.XXXXX}
|
| 33 |
+
}
|
| 34 |
+
```
|