| | --- |
| | language: |
| | - en |
| | license: mit |
| | task_categories: |
| | - text-generation |
| | - text-classification |
| | tags: |
| | - cyber-security |
| | - red-teaming |
| | - chain-of-thought |
| | - agent-reasoning |
| | - synthetic |
| | dataset_info: |
| | features: |
| | - name: instruction |
| | dtype: string |
| | - name: response |
| | dtype: string |
| | - name: hypothesis |
| | dtype: string |
| | - name: confirmed |
| | dtype: bool |
| | - name: severity |
| | dtype: string |
| | splits: |
| | - name: train |
| | num_bytes: 73400320 |
| | num_examples: 100000 |
| | download_size: 73400320 |
| | dataset_size: 73400320 |
| | configs: |
| | - config_name: default |
| | data_files: |
| | - split: train |
| | path: security_dataset.jsonl |
| | --- |
| | |
| | # 🛡️ Security Analyst CoT Dataset (100k) |
| |
|
| | A massive-scale, synthetically generated dataset designed to train **AI Security Agents** in offensive reasoning, vulnerability verification, and false positive reduction. |
| |
|
| | ## Dataset Summary |
| | - **Size:** 100,000 Unique Samples |
| | - **Format:** JSONL |
| | - **Focus:** Chain-of-Thought (CoT) Reasoning for Web Security |
| | - **Logic:** Observation -> Hypothesis -> Evidence -> Decision (O-H-E-D) |
| |
|
| | ## Features |
| | Each sample simulates a complete cognitive process of a Senior Security Analyst: |
| | 1. **Instruction**: The raw HTTP request + The server's response code/body. |
| | 2. **Reasoning**: A ` |
| |
|
| | <div class="think">` block analyzing the anomaly, evidence, and conclusion.<br />3. **Verdict**: Structured labels for `STATUS`, `SEVERITY`, and `NEXT_TEST`.<br /><br />## Attack Categories<br />The dataset covers a wide spectrum of modern web threats:<br />- **Injection**: SQLi, XSS (Reflected/Stored), Command Injection, LDAPi.<br />- **Business Logic**: Price Manipulation, Mass Assignment, Race Conditions.<br />- **Protocol**: HTTP Request Smuggling, Host Header Injection.<br />- **Anomalies**: Zero-Day simulations (Unknown patterns) and Fuzzing noise.<br />- **Benign**: High-entropy legitimate traffic to train False Positive rejection.<br /><br />## Sample Structure<br />```json<br />{<br /> "instruction": "Analyze: GET /api/v1/user?id=1' OR 1=1 HTTP/1.1...",<br /> "response": "<think>\nObservation: ...\nHypothesis: ...\nEvidence: ...\nDecision: ...\n</div> |
| | |
| | \nSTATUS: VULNERABLE...", |
| | "hypothesis": "SQL Injection", |
| | "confirmed": true, |
| | "severity": "Critical" |
| | } |
