Back to search
security-response-skills
MarketplaceWorks with Claude CodeGitHub
Security check and response skills for Claude Code and Codex: a fast read-only check, then a runbook per finding
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/nathankim0--security-response-skills)HTML
<a href="/listings/nathankim0--security-response-skills"><img src="/api/listings/nathankim0--security-response-skills/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Claude Code plugin:
claude plugin marketplace add nathankim0/security-response-skills claude plugin install security-response@security-response-skills
Or clone and link the skill folders (Claude Code and Codex; `git pull` updates them):
git clone https://github.com/nathankim0/security-response-skills.git cd security-response-skills ./install.sh # both; or: ./install.sh claude | ./install.sh codex
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 13 lines
- {
- "name": "security-response-skills",
- "owner": { "name": "nathankim0", "url": "https://github.com/nathankim0" },
- "metadata": { "description": "Security check and response skills for Claude Code and Codex", "version": "1.0.0" },
- "plugins": [
- {
- "name": "security-response",
- "source": "./",
- "description": "Fast read-only security checks plus per-finding response runbooks. First skill: NGINX2Root and Java4Shell.",
- "category": "security"
- }
- ]
- }
README.md · README · 114 lines
- <p align="center">
- <img src="assets/banner.png" alt="security-response-skills banner" width="100%">
- </p>
- <p align="center">
- <b>English</b> · <a href="README.ko.md">한국어</a>
- </p>
- # security-response-skills
- **When a critical advisory drops, find out in seconds which of your servers are affected, then fix only those, one runbook at a time.**
- ## What it is
- A collection of agent skills for Claude Code and Codex. Each skill covers one security issue. It runs a read-only check across hosts and containers, keeps only the findings that need a response, and walks the agent through the fix for each one, then checks again.
- ## Why it exists
- Advisories like NGINX2Root and Java4Shell come with long write-ups: version ranges, CPU conditions, config conditions, backported distro packages, temporary mitigations. Reading all of that while servers are exposed is slow, and it is easy to patch the wrong thing (the JDK on `PATH` instead of the one the service runs) or stop at a mitigation. These skills turn each advisory into a check script and a set of runbooks.
- ## Skills
- | Skill | Covers | Checks |
- |---|---|---|
- | [`nginx2root-java4shell`](skills/nginx2root-java4shell/SKILL.md) | NGINX2Root (CVE-2026-123007, -123633, -121810, -123634, -113425), Java4Shell (CVE-2026-114938), Linux SCTP privilege escalation | nginx version and distro backport, HTTP/2, HTTP/3, resolver, mail proxy, stale binary after upgrade, crash logs; JVM version per running process, AVX-512 VBMI CPU gate, `-XX:-UseBASE64Intrinsics`; SCTP module state |
- More skills will be added here.
- ## How every skill works
- 1. **Check**: `scripts/check.sh` prints one line per finding: `STATUS ACTION_ID TARGET DETAIL`.
- 2. **Triage**: only `ACTION` (fix needed), `MITIGATE` (temporary measure while the fix waits) and `REVIEW` (confirm by hand) rows are kept.
- 3. **Respond**: each `ACTION_ID` has a section in `references/remediation.md` with steps, verification and rollback. Patches come first; mitigations only when the patch has to wait.
- 4. **Verify**: the check runs again. Done when no `ACTION` row is left.
- ## Example
- `check.sh` run as root on a Docker host (ARM VM) with three nginx containers and one JVM, from the test run while building this skill:
- ```text
- $ sudo bash check.sh
- # nginx2root-java4shell check v1.0.0 host=colima os=Linux arch=aarch64 date=2026-11-20T01:54:56Z root=yes
- STATUS ID TARGET DETAIL
- INFO CPU colima arch=aarch64 (ARM, Java4Shell path not reachable)
- INFO NGX-PATCH colima no nginx binary found on host
- INFO JAVA-PATCH colima no Java runtime found on host (outside containers)
- ACTION NGX-PATCH colima:docker:t-nginx-old (nginx:1.24-alpine) nginx 1.24.0 is in the affected range and no backported fix was found
- MITIGATE NGX-H2-OFF colima:docker:t-nginx-old (nginx:1.24-alpine) HTTP/2 enabled (CVE-2026-123007, CVE-2026-123633)
- MITIGATE NGX-RESOLVER colima:docker:t-nginx-old (nginx:1.24-alpine) resolver with mirror/SSI subrequests or a variable proxy_pass (CVE-2026-123634)
- OK NGX-PATCH colima:docker:t-ubuntu (ubuntu:24.04) nginx 1.24.0, distro package changelog lists CVE-2026-123007 (backported fix)
- INFO JAVA-PATCH colima:docker:t-java (eclipse-temurin:21-jdk) Java 21.0.13 below fixed 21.0.13.1, but arch=aarch64 (ARM, Java4Shell path not reachable); patch on the normal cycle and recheck if the VM moves to another CPU type
- OK NGX-PATCH colima:docker:t-nginx-alpine (nginx:alpine) nginx 1.31.9 >= 1.31.9 (mainline fix)
- ACTION KRN-SCTP colima:kernel 6.8.0-117-generic sctp module can be autoloaded; block it unless SCTP is in use
- # summary host=colima ACTION=2 MITIGATE=2 REVIEW=0 OK=2 INFO=4
- ```
- The agent then works only on the `ACTION` and `MITIGATE` rows.
- ## Requirements
- - Targets: Linux with bash (macOS works for the version checks). Root gives full results; without it the script reports what it could not see.
- - Optional on targets: `docker`, `nsenter`, `kubectl`, `openssl`, `curl`.
- - An agent that loads skills: Claude Code or Codex.
- ## Install
- Claude Code plugin:
- ```bash
- claude plugin marketplace add nathankim0/security-response-skills
- claude plugin install security-response@security-response-skills
- ```
- Or clone and link the skill folders (Claude Code and Codex; `git pull` updates them):
- ```bash
- git clone https://github.com/nathankim0/security-response-skills.git
- cd security-response-skills
- ./install.sh # both; or: ./install.sh claude | ./install.sh codex
- ```
- ## Usage
- Ask the agent in plain words, for example "NGINX2Root랑 Java4Shell 점검해줘" or "check web1 web2 api1 for the nginx HTTP/2 RCE". The skill picks the targets, runs the check, shows the triage table, and confirms before changing production.
- The check script also runs on its own:
- ```bash
- sudo bash skills/nginx2root-java4shell/scripts/check.sh
- ssh web1 'sudo -n bash -s' < skills/nginx2root-java4shell/scripts/check.sh
- bash skills/nginx2root-java4shell/scripts/check.sh --k8s --probe example.com
- ```
- Exit code 1 means at least one `ACTION` or `MITIGATE` row, so it also works in CI or cron.
- ## Repository layout
- ```text
- .claude-plugin/ plugin and marketplace manifests
- skills/<name>/SKILL.md the skill: scope, check, triage, respond, verify
- skills/<name>/scripts/check.sh read-only check
- skills/<name>/references/ remediation runbooks, advisory summary, compromise checks
- tests/run.sh check script tests with fake binaries
- install.sh links skills into Claude Code and Codex
- AGENTS.md rules for agents, including how to add a skill
- ```
- ## Contributing a skill
- Follow "Adding a skill" in [AGENTS.md](AGENTS.md): one folder per issue, a read-only `check.sh` that prints the shared finding format, and one runbook section per action id.
- ## License
- MIT
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.