Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
size_categories:
|
| 6 |
+
- n<1K
|
| 7 |
+
task_categories:
|
| 8 |
+
- text-classification
|
| 9 |
+
tags:
|
| 10 |
+
- cybersecurity
|
| 11 |
+
- cloudtrail
|
| 12 |
+
- aws
|
| 13 |
+
- attack-reconstruction
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
# CybersecJSONSchemaBench — CloudTrail v6 (Leak-Free)
|
| 17 |
+
|
| 18 |
+
100 long-context CloudTrail attack-reconstruction tasks with synthetic MITRE ATT&CK chain injection into real benign flAWS logs.
|
| 19 |
+
|
| 20 |
+
**Public split:** prompt + answer_schema + records only. Gold answers and construction metadata are explicitly STRIPPED to prevent agent contamination. For scoring, see the companion gated dataset `achinta3/cybersec-v6-gold`.
|
| 21 |
+
|
| 22 |
+
**Stripped fields (vs the older `achinta3/cybersec-jsonschemabench-cloudtrail-v6`):**
|
| 23 |
+
- `record["source_file"]` — was a deterministic synthetic/benign oracle.
|
| 24 |
+
- Row-level `template_id`, `template_name`, `chain_event_count`, `phase_order`.
|
| 25 |
+
- `gold_answer`, `gold_metadata` — moved to the private gold dataset.
|
| 26 |
+
- `input_artifact.{slice_jsonl_path, path}`, `answer_schema_path` — path leakage.
|
| 27 |
+
|
| 28 |
+
**Provenance:** Built from the frozen v6.1.0 checkpoint (commit e84b3d6 of JsonSecurityBench).
|