patchman - Defensive AI security audit skill for Claude and C

项目说明

patchman

![Stars]()
![Release]()

Defensi audit skill pack for agentic code review and web application assessment.

---

O

Patchman is a skill and plugin bundle for authorized security reviews. It operates in read-first, re mode — inspecting code and architecture against OWASP and common appsec failure patterns, then producing structured findings with actionable remediation guidance.

Supported agents: Claude Code · Codex · Cursor · Windsurf · Copilot · Gemini-style skill installers

---

Installation

Claude Code (Recommended)

Restart Claude Code. The session hooks activate automatically.

installation

Uninstall

---

Claude Code — Local Install

---

Codex — Local Install

Use ~/.codex/plugins/patchman if you want to point Codex directly at the plugin path.

Use codex marketplace add ./patchman if you want Patchman to appear in Codex marketplace or picker discovery.

---

Review Modes

| Mode | Command |
|---|---|
| Full security audit | /security-audit focus=full se>=medium output=report |
| Auth review | /auth-review area=login,session,reset |
| Business logic review | /bizlogic-review feature=billing-upgrade workflow=in |
| API review | /api-review surface=public-api include=authz,rate-limit,headers |
| ORM / data-access review | /query-review path=app/models include=n-plus-one,tenant-scope |
| PR diff review | /pr-diff-review base=main head=feature/auth-refactor |
| Quick triage | /quick-triage path=admin/ reason=pre-release |
| Threat modeling | /threat-model feature=file-import |
| Audit report | /audit-report format=engineering-summary |

---

What Patchman Detects

---

Example Prompts

Full audit

Targeted review

PR review

---

Findings Format

Every finding includes:

---

Scope and Limitations

Patchman is for authorized defensi only.

In scope: secure code review, architecture review, configuration review, exploitability analysis in plain language, remediation planning

Out of scope: unauthorized intrusion, exploit weaponization, credential har, malware or persistence guidance, destructi, li chains

Patchman infers risk from static code and configuration. It does not replace runtime and will request missing deployment, proxy, or identity-boundary context when e is insufficient.

---

Repository Structure

---

Contributing

Contributions welcome. Open a PR with any of the following:

See CONTRIBUTING.md to get started.

---

License

See LICENSE.

项目信息

开发语言：JavaScript

Star 数：3

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 AI security audit skill for Claude and Codex. Reviews codebases and web apps for OWASP Top 10, auth and authorization flaws, business logic bugs, tenant isolation gaps, insecure defaults, secret leaks, and ORM or N+1 query issues, then produces clear remediation guidance.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。