Claude-Red Offensive Security Library

Problem: security analysis often degrades into fragmented Q&A

In penetration testing, vulnerability research, or bug bounty triage, engineers usually need a coherent toolchain for a specific attack surface: SQL injection, SSRF, ADCS abuse, EDR evasion, wireless protocol attacks, cloud privilege escalation, and so on. The hard part is rarely “are there tools?” It is whether the methodology is complete, the workflow is reproducible, and the boundaries are clear. If a large amount of security content is dumped into context at once, the model tends to produce generic advice and loses the depth needed for a concrete target.

How it works: loading structured modules by attack surface

Claude-Red splits security content into independently loadable SKILL.md modules. Each module organizes expert-level workflows around a clearly defined attack surface, such as:
- Web applications: SQL injection, XSS, SSRF, SSTI, XXE, RCE, deserialization, request smuggling, and business logic
- Identity and directory services: JWT attacks, OAuth abuse, Kerberoasting, ADCS ESC1-15, delegation, and persistence
- Infrastructure and red teaming: initial access, EDR evasion, Shellcode, Windows mitigations, and boundary traversal
- Exploit development: stack and heap issues, ROP chains, crash analysis, TOCTOU, and mitigations
- AI security: prompt injection, jailbreaking, and RAG poisoning

When used, start from the category index, locate the relevant attack surface, and let the model load the matching module based on conversation triggers. This turns generic chat into target-specific structured analysis, reducing irrelevant context overhead.

Scope and cautions

This material is intended for authorized red-team exercises, security research, CTF preparation, bug bounty triage, and training scenarios. The assumptions include:
- the target asset must be explicitly within the authorized scope
- the selected module must match the current attack surface
- outputs should reflect security boundaries, legal constraints, and test evidence
- report formatting can follow standards such as CVSS, but this does not replace full human review