Run external attack surface analysis and red teaming with Claude Code
Testing websites or APIs often requires more than simple chat; it demands the perspective of an expert following real attack scenarios. This repository teaches Claude Code the skills and reporting discipline needed for external attack surface analysis and red teaming. It is a skill bundle that organizes patterns for 24 vulnerability classes based on 681 public HackerOne reports.
Four layers from thinking to delivery
The skill bundle is structured into four layers: thinking, web app hunting, peripheral infrastructure impact, and result delivery. It first teaches a 5-step non-linear workflow and the mindset of a red team operator, then applies specific patterns like XSS or SQL injection through 58 web app vulnerability skills. It subsequently covers the latest CVE chains and post-credential theft escalation techniques for enterprise platforms such as M365, Okta, and vCenter. Finally, it writes reports only for verified results that pass a 7-question gate. All skills load automatically based on the topic when the user describes the test target in plain English.
Installation paths and supported agents
You can install it from the Claude Code plugin marketplace or clone the repository and copy-install it via script. The plugin method maintains namespaces and makes updates easy, while the copy-install method allows using the cbh CLI and hunt scaffolder together. This skill works in the same SKILL.md format across various agent harnesses, including Claude Code, OpenCode, OpenAI Codex CLI, Hermes Agent, and Google AntiGravity. However, slash commands and the /hunt engine are designed exclusively for Claude Code.
Design focused solely on external attack surfaces
This repository handles external attack surfaces directly accessible from the internet, intentionally excluding internal Active Directory attacks or the use of C2 frameworks. It supports bug bounty hunting, web application penetration testing, external red team engagements, cloud misconfiguration, reconnaissance and OSINT, and report writing tailored to various platforms. Internal network penetration or post-exploitation stages must be combined with other tools, as this repository is clearly separated at that boundary.
Things to check before use
This skill is designed to be used only on assets you own or have written authorization for. Claude models have cyber safety mechanisms that may refuse approved offensive security tasks, so if you have official authorization, it is recommended to join Anthropic's Cyber Verification Program (CVP). Additionally, some high-risk requests in the Opus 5 model may be switched to Opus 4.8 due to the model's safety policy, so you should carefully watch for model switching notifications during sessions. The license is dual-licensed with MIT for code and CC BY 4.0 for content, requiring attribution for commercial reuse.