claude-red: 78 Offensive Security Skills for Claude Code and Claude

Claude Skills are a simple idea: a folder with a SKILL.md file that Claude loads on demand when your conversation matches its description. claude-red applies that idea to offensive security. It’s an open-source library of 78 skills, one per attack surface, designed to make Claude behave like a methodical red team operator rather than a generic chatbot that gives vague answers about SQL injection.

The repository is SnailSploit/Claude-Red, MIT licensed, and at the time of writing it has close to 10,000 GitHub stars. This post covers what’s in it, how to install it, and how to use it sensibly. I’ve read through the README and skill index rather than running every skill, so treat this as an overview, not a tested review of each one.

What claude-red is

Each skill is a structured SKILL.md that primes Claude with methodology for one area: techniques, tooling, edge cases and escalation paths. Because skills load on demand from conversational triggers, you don’t pay context for the ones you aren’t using. Mention SQL injection and Claude pulls in offensive-sqli; ask about Kerberoasting and the Active Directory skill loads instead.

The README lists 78 skills across 23 categories. The project says it’s meant for authorised red team engagements, bug bounty triage, security research, CTF preparation and operator training.

What’s covered

• Web application: the largest group, with SQL injection, XSS, SSRF, SSTI, XXE, IDOR, file upload, deserialization, race conditions, request smuggling, GraphQL, WAF bypass and business logic.
• Auth and identity: JWT and OAuth/OIDC attacks.
• Wireless: 14 skills covering Wi-Fi recon, WPA2 and WPA3, WPA Enterprise, WPS, evil twin, Bluetooth, Zigbee, Z-Wave and LoRa.
• Active Directory, cloud, mobile and IoT: a skill each for on-prem AD, multi-cloud attack paths, Android and iOS testing, and IoT and embedded work.
• Infrastructure and red team: initial access, advanced red team operations, Windows internals and related areas.
• Exploit development and fuzzing: stack and heap corruption, mitigations, crash analysis, libFuzzer and AFL++.
• Everything else: API security, containers and Kubernetes, CI/CD pipelines, cryptography, privilege escalation, post-exploitation, supply chain, network attacks, AI security (prompt injection, RAG poisoning), OSINT, and utility skills for triage checklists and professional reporting.

How to install it

The recommended route is to clone it into your Claude skills folder:

git clone https://github.com/SnailSploit/claude-red ~/.claude/skills/claude-red

If you only want a couple of categories, a sparse checkout keeps the footprint small:

git clone --filter=blob:none --sparse https://github.com/SnailSploit/claude-red
cd claude-red && git sparse-checkout set Skills/web Skills/active-directory

There’s also an install.sh script that takes a target folder or a single category, for example ./install.sh --category web. On Claude.ai you can paste a SKILL.md into a Project’s instructions, and with Claude Code you can feed a single skill in as a system prompt.

Skills worth trying first

• Triage and reporting: the utility skills for fast checklists and professional write-ups are the safest and most immediately useful, whatever your role.
• API security and web skills: useful for bug bounty and for testing your own applications against the OWASP lists.
• Wireless skills: a good companion if you already use Kali for Wi-Fi assessments on your own lab network.
• Fuzzing and crash analysis: handy for learning how bugs are found and judged for exploitability.

Use it responsibly

These skills cover offensive techniques, including evasion, post-exploitation and social engineering. Only use them against systems you own or have written permission to test, and keep to your agreed scope. Skills don’t give Claude new abilities or tools; they shape how it reasons and what it recommends, and Claude may still decline requests it judges harmful.

Two practical cautions:

• Read a skill before you load it. A SKILL.md is instructions your AI will follow, so a third-party repo is a supply-chain risk like any other. Skim the files you install and pin to a commit you’ve reviewed.
• Keep approvals on. If you run these skills in Claude Code, leave command approvals enabled so nothing runs against a target without your say-so.

Where it fits

claude-red is a knowledge layer, not a scanner. You still need real tools such as Burp, nmap and the aircrack-ng suite to do the testing, and Claude with these skills acts as a methodical assistant for planning, interpreting results and writing them up. If you’re building out your Claude setup more broadly, my Claude MCP hub collects the servers, skills and plugins I’ve covered so far.