How do I install the Kernel Security Audit skill?

Run npx skillmds@latest add gvago/kernel-security-audit in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Kernel Security Audit safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Kernel Security Audit?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Kernel Security Audit free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under Apache-2.

Who published Kernel Security Audit?

gvago (@gvago) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Kernel Security Audit by gvago — Agent Skill

Name: Kernel Security Audit
Author: gvago

Kernel Security Audit

Use when a Linux kernel PR diff handles untrusted input (copy_from_user, get_user, ioctl args, netlink attrs, network or firmware data), security lens (Sashiko stage 6): OOB access, integer overflow, TOCTOU, info leaks to userspace, privilege escalation in the CHANGED code only. Part of the linux-kernel-review suite.

gvago 7af6aa3 2.5 KB Updated Sep 17, 2026

File contents

kernel-security-audit

Lens for security review of a kernel patch (Sashiko stage 6). Red Team posture: scrutinize every point where untrusted input reaches sensitive code.

How to apply

Map the data boundaries first: what enters from userspace (copy_from_user, get_user, ioctl args, netlink attrs, syscall params), from the network, or from untrusted hardware/firmware.
Follow each untrusted value to every use. A missing or weak validation between entry and use is the finding.
Reachability by untrusted, remote, or unprivileged input RAISES severity; believed unreachability never lowers it (see kernel-review-discipline).

Rule index

S1. Buffer overflows and out-of-bounds reads/writes: index arithmetic on attacker-influenced values, missing or off-by-one bounds checks, memcpy/strcpy with unchecked lengths.
S2. Integer overflow/underflow feeding allocation sizes, offsets, or loop bounds; truncating casts that bypass checks done at the wider width.
S3. TOCTOU races: validation and use in separate windows where the value can change (re-fetch from userspace, racing state change).
S4. Information leaks to userspace: copy_to_user of structs with uninitialized fields or padding holes; error paths leaking kernel pointers or heap content.
S5. Privilege escalation vectors: capability checks missing or performed on the wrong credential; state changes ordered so an unprivileged step gains a privileged effect.

Mapping a finding to the contract

action_level: action_required for any proven S1-S5 with an untrusted path; remediation_recommended when the input source is privileged but the pattern is still unsafe.
category: Security.
evidence: entry point of the untrusted value, the propagation chain, the unsafe use, functions and lines. State the attacker precondition.

What NOT to flag

Bounds checks on values provable to originate only from trusted kernel state.
Constructs that look unsafe but sit behind capability checks you can cite, cite them and dismiss instead.
Hardening suggestions (e.g. "use strscpy everywhere") without a proven vulnerable path.

Sourcing

Stage text adapted from Sashiko (Apache-2.0).

---name: kernel-security-auditdescription: Use when a Linux kernel PR diff handles untrusted input (copy_from_user, get_user, ioctl args, netlink attrs, network or firmware data), security lens (Sashiko stage 6): OOB access, integer overflow, TOCTOU, info leaks to userspace, privilege escalation in the CHANGED code only. Part of the linux-kernel-review suite.license: Apache-2.0---# kernel-security-auditLens for security review of a kernel patch (Sashiko stage 6). Red Teamposture: scrutinize every point where untrusted input reaches sensitivecode.## How to apply- Map the data boundaries first: what enters from userspace (copy_from_user, get_user, ioctl args, netlink attrs, syscall params), from the network, or from untrusted hardware/firmware.- Follow each untrusted value to every use. A missing or weak validation between entry and use is the finding.- Reachability by untrusted, remote, or unprivileged input RAISES severity; believed unreachability never lowers it (see kernel-review-discipline).## Rule index- S1. Buffer overflows and out-of-bounds reads/writes: index arithmetic on attacker-influenced values, missing or off-by-one bounds checks, memcpy/strcpy with unchecked lengths.- S2. Integer overflow/underflow feeding allocation sizes, offsets, or loop bounds; truncating casts that bypass checks done at the wider width.- S3. TOCTOU races: validation and use in separate windows where the value can change (re-fetch from userspace, racing state change).- S4. Information leaks to userspace: copy_to_user of structs with uninitialized fields or padding holes; error paths leaking kernel pointers or heap content.- S5. Privilege escalation vectors: capability checks missing or performed on the wrong credential; state changes ordered so an unprivileged step gains a privileged effect.## Mapping a finding to the contract- action_level: action_required for any proven S1-S5 with an untrusted path; remediation_recommended when the input source is privileged but the pattern is still unsafe.- category: Security.- evidence: entry point of the untrusted value, the propagation chain, the unsafe use, functions and lines. State the attacker precondition.## What NOT to flag- Bounds checks on values provable to originate only from trusted kernel state.- Constructs that look unsafe but sit behind capability checks you can cite, cite them and dismiss instead.- Hardening suggestions (e.g. "use strscpy everywhere") without a proven vulnerable path.## SourcingStage text adapted from Sashiko (Apache-2.0).

Frequently asked questions