snailsploit[$]
⌘K live
claude-red
skills library
mit license
38 SKILL.md files
9 categories
Claude Skills system

claude-red.
38 offensive skills.

A curated library of offensive-security skills for the Claude Skills system. SQLi, shellcode, EDR evasion, exploit dev — 38 SKILL.md files spanning the work an offensive operator actually does. Drop-in capabilities for agent harnesses that need real adversarial coverage. Each skill is small, sharp, and reads in under a minute.

01 · categories
Nine categories, 38 skills. Names are self-describing — no marketing, just craft.

9 categories.

Web exploitation
6 skills

SQLi · SSRF · CSRF · path traversal · prototype pollution · XSS chains

Binary exploitation
5 skills

Stack overflows · heap shaping · ROP · format strings · ret2libc

Shellcode & payloads
4 skills

x86_64 / arm64 / msf-compatible · staged · stageless · custom encoders

EDR evasion
4 skills

Direct syscalls · unhooking · AMSI bypass · ETW patching

Network & C2
5 skills

C2 channel design · DNS tunneling · ICMP exfil · domain fronting · jitter

Cloud & container
4 skills

K8s breakouts · IMDS abuse · IAM enumeration · privesc paths

AD & windows
3 skills

Kerberoasting · DCsync · BloodHound triage

Linux post-ex
3 skills

Persistence · privesc · lateral · log evasion

Recon & osint
4 skills

Subdomain · CT · cloud bucket discovery · code-search recon

02 · format
Why each skill is a single file, why it's that short, why it's not a script.

one skill, one file.

Each skill is a single SKILL.md file. The file is short — usually 80 to 200 lines — because skills aren't scripts. They are operating instructions for an agent that will write the script, in context, against a target the agent has in front of it.

The skill describes when to apply the technique, what to look for, what the expected indicators of success are, and what to do if it doesn't land. It does not pre-bake the exploit, because the exploit only makes sense in context. Pre-baking is what makes the average "AI security skill" library brittle the moment the target deviates from the demo.

This is also why Claude-Red is small. 38 skills, not 380. Most "more" is filler — and a skill an operator can't memorize the shape of is a skill the operator won't reach for.

03 · usage
How to wire a Claude-Red skill into an agent harness in three steps.

how to use it.

# 1. clone
$ git clone https://github.com/SnailSploit/claude-red

# 2. point your skills loader at it
$ export CLAUDE_SKILLS_DIR=$PWD/claude-red/skills

# 3. drop into a harness
The agent now has 38 new tools available, all reachable through
the standard SKILL.md discovery path. No code changes required.
more frameworks all frameworks →
AATMF →Adversarial AI threat modelingSEF →Social engineering frameworkP.R.O.M.P.T →Compositional grammarToolkit →LLM safety CLIPlaybook →Diagnostic methodology