Doc-foundry是一款利用大语言模型(LLM)自动生成并验证网络安全Sigma检测规则的工具 1。该工具采用“代理提出,引擎证明”的核心架构,将创造性工作交给LLM,而真理判定则由确定性代码负责 1。这一设计通过分离“提出者”与“证明者”的角色来执行对抗循环,旨在确保证据的可复现性和诚实性 1。
在技术实现上,Doc-foundry包含四个角色特定的LLM代理:威胁分析师、蓝队检测工程师、遥测系统以及红队对手 1。其中,红队代理在首轮即承诺使用两个真实记录的攻击,并在后续轮次中重复使用该攻击集,以防止移动球门柱 1。系统的确定性裁判组件负责计算假阳性、假阴性、对抗规避能力以及覆盖率标签(如real-coverage、brittle、broken) 1。此外,该工具支持离线运行,依赖Atomic Red Team作为正样本库,并以MITRE ATT&CK和LOLBAS作为知识基础 1。
Doc-foundry提供两条主要流水线以优化规则的可用性与正确性 1。一是Proof Pipeline,用于通过严格测试证明规则的有效性;二是Review Chain,模拟SOC主管和红队的反馈流程,以调整告警量和误报率 1。示例运行展示了该工具在处理CVE-2026-48939 (Joomla iCagenda) 和 T1218.011 (rundll32) 时的表现 1。在部署环节,该工具通过Kartikeya's Gate进行治理,计算影响范围与能力差异,并强制要求获得人类批准后方可部署 1。
Doc-foundry is a tool that leverages large language models (LLMs) to generate and validate Sigma security detection rules, employing an architecture where LLM agents propose rules and deterministic engines prove their efficacy 1. The system separates creative generation from truth verification by using four specialized LLM roles—Threat Analyst, Blue Detection Engineer, Telemetry, and Red-Team Adversary—to simulate a rigorous adversarial cycle 1. A deterministic judge, implemented in engine.py and evaluator.py, calculates false positives, false negatives, adversarial evasion, and coverage metrics, assigning labels such as "real-coverage," "brittle," or "broken" based on objective analysis 1.
The platform operates through two primary pipelines designed to ensure both technical correctness and operational usability 1. The Proof Pipeline verifies rule validity through strict testing, while the Review Chain simulates feedback from stakeholders like SOC leads and red teams to optimize alert volumes and reduce false positive rates 1. To prevent shifting goalposts, the Red-Team agent commits to two real recorded attacks in the first round and reuses this same attack set in subsequent iterations 1. Example executions of the tool have addressed threats including CVE-2026-48939 related to Joomla iCagenda and MITRE ATT&CK technique T1218.011 involving rundll32 1.
Deployment governance for Doc-foundry relies on a mechanism called "Kartikeya's Gate," which computes a Blast Radius multiplied by a Capability Delta metric before allowing release 1. This process strictly requires human approval prior to deployment, ensuring accountability despite the automated generation 1. The tool supports offline operation and utilizes Atomic Red Team as its positive sample library, drawing knowledge foundations from MITRE ATT&CK and LOLBAS frameworks 1.
评论
还没有评论,欢迎留下第一条。