claude-red is a curated library of offensive security skills for the Claude Skills system, packaged as drop-in SKILL.md files that prime Claude with expert methodology for specific attack surfaces. It is built for red teamers, penetration testers, bug bounty hunters, and security researchers who run authorized engagements and want the model to behave like a specialist rather than a generalist when a technical topic comes up.
What it does
The project ships a large collection of skills — the README advertises 78 skills across 23 categories — each aimed at one slice of offensive work. Instead of describing a technique in the abstract, a skill loads a structured methodology: the relevant techniques, the tooling, the edge cases, and the escalation paths for that surface. Coverage runs wide. The materials call out SQL injection, shellcode, EDR evasion, and Active Directory Certificate Services abuse as examples, and the broader set spans wireless, cloud, Active Directory, and exploit development. In practice that means the library reads less like a set of scripts and more like a shared playbook that a whole red team could lean on, so no single operator has to hold expert-level knowledge of every attack surface in their head at once.
The stated use cases are deliberately scoped: authorized red team engagements, bug bounty triage, security research, CTF preparation, operator training, and methodical exploration of an attack surface. These are the situations where having a domain specialist on tap saves time — you are already working within a defined scope and want faster, better-informed recall of how a particular class of attack actually plays out.
How it works
Each skill is a single SKILL.md file written to the format the Claude Skills system understands. Rather than loading everything up front, skills activate on demand based on conversational triggers. Mention SQL injection and the offensive-sqli skill loads; move on to another topic and a different skill takes over. The design point is that you do not spend context budget on skills you are not using, which keeps the model focused and lets the library grow without every skill competing for attention in the same conversation.
Once a skill is active, Claude effectively adopts that domain's expertise for the duration of the exchange. The framing throughout the materials is that Claude becomes an on-demand, context-aware operator: it knows the moves for the surface in front of it and can walk through the methodology step by step as the engagement unfolds.
Getting started
Installation is a single clone into the Claude skills directory:
git clone https://github.com/SnailSploit/claude-red ~/.claude/skills/claude-red
After that, Claude auto-loads matching skills from your conversation, so there is no separate configuration step to wire up each one. The README also points to installing a single category rather than the whole library when you only need a narrow slice, though the exact command is cut off in the saved copy. The repository is MIT licensed and its main language is Python, and it has drawn a strong following — over 3,500 stars and several hundred forks — which is a reasonable signal that the format resonates with practitioners.
When to use it / when not
Reach for claude-red when you are inside an authorized scope and want faster access to deep methodology: a live red team engagement, a bug bounty program where you are triaging findings, a CTF, or a training exercise where an operator is learning a surface. It is also a natural fit for research where you want a structured, expert-level starting point instead of assembling one yourself.
It is not a substitute for permission. Everything here is offensive tooling, and the materials frame its use as authorized work — using it against systems you do not have explicit approval to test is out of bounds and likely illegal. It is also worth remembering that these skills encode methodology and judgment, not push-button automation; they make Claude a better-informed collaborator, but the operator still owns the decisions and the consequences.
Anyone running legitimate offensive security work should take this repository seriously. It packages hard-won methodology into a format that loads exactly when it is relevant, spreads specialist knowledge across a team without forcing everyone to be an expert in everything, and installs in a single command. The value is real for authorized red teams, bounty hunters, and researchers; the responsibility that comes with it is equally real, and the project is only worth adopting by people prepared to stay firmly on the authorized side of that line.