claude-security-skills - Defensive security skills for Claude Code and the

项目说明

<di="center">

Claude Security Skills

English · 日本語 · 简体中文

Production-grade defensive and offensive security skills for Claude Code and the Claude Agent SDK.


</div>

---

What this is

A curated set of Claude Skills — opinionated, playbooks Claude loads on demand — for hardening modern web applications and generative-AI systems.

Two complementary halves:

Out of scope: red-team tooling for third-party systems, exploit de, e of detection, mass scanning.

Skill catalog

Defensive — code review & scoring (skills/defensive/)

| Framework / SDK | Skill | Detects |
|---|---|---|
| Next.js | nextjs-security-scan | en, ser, middleware, CORS, CSP |
| Express | express-security-scan | middleware order, helmet, jwt, sendFile tra |
| Django / DRF | django-security-scan | DEBUG, raw ORM, mark_safe, AllowAny, fields="__all__" |
| Spring Boot | spring-boot-security-scan | permitAll, CSRF, JdbcTemplate, Jackson, actuator |
| FastAPI | fastapi-security-scan | Depends, JWT, Pydantic, SQL, path tra |
| NestJS | nestjs-security-scan | , guards, DTOs, TypeORM raw |
| OpenAPI | openapi-spec-security-scan | global security, schema tightness, response leaks |
| Anthropic SDK | anthropic-sdk-security-scan | prompt injection, tool input, prompt-cache PII |
| OpenAI SDK | openai-sdk-security-scan | function args, structured output, Assistants threads |
| AI SDK | can | tool execute, attachments, streamText XSS |
| LangChain | langchain-security-scan | REPL/Shell, RAG trust, callbacks |
| MCP server | mcp-ser | tool fs/exec/SSRF, transport auth |

Findings format: skills/SCORING.md.

Offensive — self-pentest of your own app (skills/offensive/)

| Framework / SDK | Skill | Probes |
|---|---|---|
| *(any)* | webapp-pentest-checklist | OWASP Web/API Top 10 baseline |
| Express | express-attack-probe | prototype pollution, HPP, trust-proxy spoofing |
| Django | django-attack-probe | DEBUG leak, host injection, DRF AllowAny, mass-assign |
| Spring Boot | spring-boot-attack-probe | actuator, h2-console, JWT confusion, mass-assign |
| Next.js | nextjs-attack-probe | NEXT_PUBLIC leak, middleware bypass, ser auth |
| FastAPI | fastapi-attack-probe | OpenAPI enum, Depends gaps, Pydantic extra fields |
| NestJS | nestjs-attack-probe | , guard misses, ws auth |
| *(any)* | prompt-injection-probe | direct/indirect/multi-turn injection battery |
| Anthropic SDK | anthropic-sdk-attack-probe | XML tag confusion, prefill abuse, tool input |
| OpenAI SDK | openai-sdk-attack-probe | function args, structured-output bypass, threads |
| AI SDK | be | tool execute, useChat auth, markdown XSS |
| LangChain | langchain-attack-probe | REPL/Shell, RAG injection, shared memory |
| MCP server | mcp-ser | path tra, transport auth, SSRF, DNS rebinding |

Probing rules: skills/PROBING.md.

Quick start

Claude Code

Pull the skills you need into your project (recommended) or user scope:

Claude Code auto-discovers the skill on next launch. Invoke with /<skill-name> or let Claude trigger it when its description matches the task.

Claude Agent SDK

Mount the skills/ directory as a skill source per the Agent SDK docs.

Authorization for offensi

Every offensi follows skills/PROBING.md:

If you can't satisfy these, the skill returns a single PREFLIGHT-BLOCKED finding and stops.

Contributing

PRs welcome. See CONTRIBUTING.md. In short, a skill is ready to merge when it:

1. Solves a concrete defensive or offensive task — not a generic security o.
2. Has a description precise enough for Claude to self-select it.
3. Includes at least one wrong- code example.
4. Cites authoritati (OWASP, NIST, docs, CVE).

License

MIT © Dolphin LLC.

Maintainer

Dolphin LLC

项目信息

开发语言：未知

Star 数：1

技术要点

该项目属于防护思路方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕Defensive security skills for Claude Code and the Claude Agent SDK — web applications and generative AI systems.展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。