Skip to main content

windows-kernel-security

Analyze Windows kernel security mechanisms used by drivers and game protection. Use for callbacks, IRQL, kernel memory, PatchGuard, DSE, HVCI, or driver trust; use dma-attack-techniques for PCIe DMA.

Source facts

Repository
NeverSight/NeverD
Last source activity
September 13, 2026 at 11:12
Detected SKILL.md language
English
Stars
95
Forks
17

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer
6 files

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
name
windows-kernel-security
description
Analyze Windows kernel security mechanisms used by drivers and game protection. Use for callbacks, IRQL, kernel memory, PatchGuard, DSE, HVCI, or driver trust; use dma-attack-techniques for PCIe DMA.
# Windows kernel security Use this skill for authorized Windows kernel and defensive game-security work. Match implementation details to the target Windows version and verify undocumented structures rather than treating historical layouts as stable. ## Topic routing Read only the reference that matches the task: - [Kernel foundations](references/kernel-foundations.md) for objects, symbols, PatchGuard, DSE, VBS, HVCI, and Secure Boot. - [Drivers and observation](references/drivers-and-observation.md) for callbacks, driver structure, hooking, APCs, and ETW. - [Memory and forensics](references/memory-and-forensics.md) for segment heap, pool artifacts, virtual or physical memory, and MDLs. - [Threats and virtualization](references/threats-and-virtualization.md) for vulnerable drivers, boot threats, PatchGuard research, and hypervisor defenses. - [Resource map](references/resources.md) for tools and upstream references. Use `anti-cheat-systems` for product-level detection architecture and `dma-attack-techniques` for PCIe/IOMMU threat modeling. Separate documented APIs from version-specific internals and label unsupported techniques accordingly.
View on GitHub