woa9875/Brahmastra-DarkNetra

Brahmastra DarkNetra 是一个大规模的开源ChatML数据集，专门用于训练AI驱动的安全扫描器和网络安全助手。数据集名称源自梵语，Netra意为眼睛或视野，DarkNetra象征着能够发现他人忽略的安全漏洞。该数据集包含838,679个样本，分为训练集（813,519个样本）和测试集（25,160个样本），格式为ChatML（消息数组）。数据集涵盖多个网络安全相关类别，包括CVE知识（346,212个样本）、渗透测试方法（318,809个样本）、漏洞利用知识（94,870个样本）、合成DAST（49,500个样本）、网络安全推理（31,853个样本）、漏洞赏金（10,432个样本）、扫描器智能（10,261个样本）、攻击方法（6,590个样本）、工具知识（5,784个样本）、攻击知识（4,426个样本）、OWASP标准（3,976个样本）和实验室报告（1,203个样本）。数据来源于多个安全资源，如Pentest-Agent、All-CVE-Records、ExploitDB等，许可证为CC BY-NC-SA 4.0，但每个样本可能包含特定的源许可证。数据集旨在用于教育和防御目的，包含漏洞利用技术描述、安全测试载荷和攻击方法，但已进行净化处理以减少误报。Brahmastra DarkNetra is the largest open-source ChatML dataset for training AI-powered security scanners and cybersecurity assistants. Named from Sanskrit: Netra = eye, vision, with DarkNetra symbolizing the ability to see vulnerabilities that others miss. The dataset contains 838,679 samples, split into 813,519 training samples and 25,160 test samples, formatted in ChatML (messages array). It covers multiple cybersecurity-related categories, including CVE knowledge (346,212 samples), penetration testing methodology (318,809 samples), exploit knowledge (94,870 samples), synthetic DAST (49,500 samples), cybersecurity reasoning (31,853 samples), bug bounty (10,432 samples), scanner intelligence (10,261 samples), attack methodology (6,590 samples), tool knowledge (5,784 samples), attack knowledge (4,426 samples), OWASP standards (3,976 samples), and lab writeups (1,203 samples). Data is sourced from multiple security resources such as Pentest-Agent, All-CVE-Records, ExploitDB, etc., under the CC BY-NC-SA 4.0 license, with individual samples having source-specific licenses. The dataset is intended for educational and defensive purposes, containing descriptions of vulnerability exploitation techniques, security testing payloads, and attack methodologies, with raw exploit payloads sanitized to reduce false positives while preserving educational value.