reverse-skills-max: initial public release (restored, redacted, hardened)
This commit is contained in:
@@ -0,0 +1,87 @@
|
||||
# 社区安全 Skill 生态对照(2026-07)
|
||||
|
||||
> 来源检索日期:**2026-07-17**
|
||||
> 目的:让 reverse-skill **知道外面有什么**,按需借鉴,**不**把外部巨型库整仓并入本包。
|
||||
> 本包身份:路由 + 工具自举 + 证据/scope 契约 + field-journal(见 `ops/IDENTITY.md`)。
|
||||
|
||||
## 1. 外部高价值仓库(可学习,勿盲装)
|
||||
|
||||
| 仓库 | 规模/定位 | 对本包价值 | 风险 |
|
||||
|------|-----------|------------|------|
|
||||
| [trailofbits/skills](https://github.com/trailofbits/skills) | ToB 安全研究 Claude 插件市场 | 审计/漏洞分析/RE 插件质量标杆 | 需按 ToB 市场安装;勿默认信任非 curated 副本 |
|
||||
| [trailofbits/skills-curated](https://github.com/trailofbits/skills-curated) | 已审核插件列表 | 优先于任意社区 skill | 同上 |
|
||||
| [Orizon-eu/claude-code-pentest](https://github.com/Orizon-eu/claude-code-pentest) | 6 个 pentest 生命周期 skill + 纯 Python 脚本 | 侦察→利用→报告流水线可对标我们 `attack-chain`+`pentest-tools` | 授权边界要自检;脚本需沙箱 |
|
||||
| [trilwu/secskills](https://github.com/trilwu/secskills) | 16 skills + 6 专家 subagent | 多角色分工可对标 `ops/role-map.md` | 插件形态,与本包 monorepo 不同 |
|
||||
| [Masriyan/Claude-Code-CyberSecurity-Skill](https://github.com/Masriyan/Claude-Code-CyberSecurity-Skill) | ~15–19 领域 skill(含 RE/OT/CSOC) | 领域覆盖 checklist | 深度不如本包单域 skill |
|
||||
| [mukul975/Anthropic-Cybersecurity-Skills](https://github.com/mukul975/Anthropic-Cybersecurity-Skills) | **800+** skill · ATT&CK/NIST 映射 | **框架映射**与领域目录可参考,不宜整库依赖 | 体量过大、维护与投毒面巨大 |
|
||||
| [Eyadkelleh/awesome-skills-security](https://github.com/Eyadkelleh/awesome-claude-skills-security) | SecLists 打包为 agent skills | 字典/payload 入口 | 与 seclists bootstrap 重叠 |
|
||||
| [securityfortech/awesome-security-skills](https://github.com/securityfortech/awesome-security-skills) | 安全 skill 精选列表 | 发现新 skill 的索引 | 列表型,需逐个审计 |
|
||||
| [VoltAgent/awesome-agent-skills](https://github.com/VoltAgent/awesome-agent-skills) | 1000+ 跨厂商 skill 索引 | 发现官方/社区 skill | 非安全专用 |
|
||||
| [anthropics/claude-code-security-review](https://github.com/anthropics/claude-code-security-review) | PR 安全审查 GitHub Action | 可对标我们 docs/报告侧「变更审计」场景 | CI 产品,不是 RE 路由 |
|
||||
| [agentskills.io](https://agentskills.io) | Agent Skills 开放标准 | frontmatter/目录约定对齐 | 标准本身无攻防内容 |
|
||||
|
||||
### 1.1 第二轮检索增补(2026-07-17 再搜)
|
||||
|
||||
| 仓库 / 资源 | 定位 | 本包落点 |
|
||||
|-------------|------|----------|
|
||||
| [trailofbits/skills](https://github.com/trailofbits/skills) 插件:`audit-context-building` `differential-review` `semgrep-rule-creator` `sharp-edges` `dwarf-expert` `burpsuite-project-parser` | 审计上下文、差分安全审查、危险 API、DWARF、Burp 工程解析 | 对照 `ida-reverse`/`docs-generator`/审计工作流;**不**整库并入 |
|
||||
| [HexRaysSA/ida-claude-code-plugins](https://github.com/HexRaysSA/ida-claude-code-plugins) | 官方 IDA Claude 插件(含 domain 自动化,标注 unsafe) | `ida-reverse` MCP 路径对照;unsafe 插件默认不启用 |
|
||||
| [P4nda0s/reverse-skills](https://github.com/P4nda0s/reverse-skills) | IDA-NO-MCP:导出反编译后再分析;rev-frida/dex-dump/u3d | 与「MCP 不可用时的离线导出」互补 |
|
||||
| [2389-research/binary-re](https://github.com/2389-research/binary-re) | triage→static(r2/Ghidra)→dynamic(QEMU/GDB/Frida)→synthesis | `reverse-engineering` 阶段门闩见 `re-agent-workflow.md` |
|
||||
| [incogbyte/android-reverse-engineering-claude-skill](https://github.com/incogbyte/android-reverse-engineering-claude-skill) | APK 解包、端点提取、自适应 Frida 绕过 | 对照 `apk-reverse`;动态脚本需 scope |
|
||||
| [OwenPawl/cerberus-re-skill](https://github.com/OwenPawl/cerberus-re-skill) | Apple 向 Ghidra+LLDB+Frida 三循环 | 可参考 macOS/iOS 动态环 |
|
||||
| [ljagiello/ctf-skills](https://github.com/ljagiello/ctf-skills) | CTF reverse/pwn;工具按需装 | 对照 CTF-Sandbox + `pwn-chain` |
|
||||
| [shuvonsec/claude-bug-bounty](https://github.com/shuvonsec/claude-bug-bounty) | /recon→/hunt→/validate→/report | 对照 `recon-pipeline.md` + scope 门 |
|
||||
| [PayloadsAllTheThings](https://github.com/swisskyrepo/PayloadsAllTheThings) | Web payload + Prompt Injection 章节 | `pentest-tools/payloads` 优先;LLM 见 `llm-security` |
|
||||
| [HackTricks](https://hacktricks.wiki/) | 渗透方法论 + **AI/MCP 滥用** | 见 skill-supply-chain MCP 节 |
|
||||
| [appsecsanta AI pentesting agents 2026](https://appsecsanta.com/research/ai-pentesting-agents-2026) | 39+ 开源 AI 渗透 agent 架构分类 | 多 agent ≠ 必须;我们用 role-map |
|
||||
| Snyk 评测「更多 skill ≠ 更好」 | 技能堆叠可能降低审计质量 | 强化「深 skill + 路由」策略 |
|
||||
|
||||
## 2. 安全标准与威胁(2025–2026)
|
||||
|
||||
| 来源 | 要点 | 本包落点 |
|
||||
|------|------|----------|
|
||||
| [OWASP Agentic Skills Top 10](https://owasp.org/www-project-agentic-skills-top-10/) | 恶意 skill、供应链、权限滥用、记忆投毒等 | `ops/skill-supply-chain.md` |
|
||||
| [Anthropic Agent Skills 工程文](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills) | 只装可信源;审脚本与依赖 | 同上 + bootstrap 禁止猜路径 |
|
||||
| ClawHavoc 等投毒活动(AST10 记载) | 注册表批量恶意 skill | 禁止从未知 registry 一键安装进本包 |
|
||||
|
||||
## 3. 本包已有 vs 外部「广」覆盖
|
||||
|
||||
| 领域 | reverse-skill | 外部常有、我们不整库并入的原因 |
|
||||
|------|---------------|--------------------------------|
|
||||
| APK/JS/IDA/r2/固件/pwn | **深** skill + 脚本 | 保持深度与 tool-index 绑定 |
|
||||
| 渗透/攻击链/SRC | pentest-tools + attack-chain + src-hunter | Orizon 类可作方法论对照 |
|
||||
| LLM/Agent 安全 | llm-security | AST10 增强 skill 自身安全 |
|
||||
| 证据/scope/角色 | **ops/**(特色) | 多数 skill 包没有 case 契约 |
|
||||
| OT/ICS / 纯 GRC / 欺诈 F3 | 无独立 skill | 路由未命中 → 提议新增或外链,不硬塞 |
|
||||
| 800+ 微 skill | 不复制 | 用 MASTER 路由 + 域 skill 替代碎片化 |
|
||||
|
||||
## 4. 借鉴规则(MUST)
|
||||
|
||||
```text
|
||||
1. 禁止 git submodule 拉整库 800+ skills 当运行时依赖
|
||||
2. 借鉴时:提取「阶段/清单/命令模式」写入本包 references 或现有 skill
|
||||
3. 外部脚本:先在隔离环境看依赖与网络行为,再考虑 bootstrap-manifest
|
||||
4. 新场景:走 CONTRIBUTING 新增 skill,并更新 routing + RULES 关键词
|
||||
5. 标注来源 URL + 检索日期(本文件格式)
|
||||
6. 安装/合并前走 ops/skill-supply-chain.md 清单
|
||||
7. 运行时只加载 MASTER-ROUTING 的 PRIMARY(+必要 secondary),避免技能堆叠过载
|
||||
```
|
||||
|
||||
## 4.1 本包已沉淀的「借鉴产物」(非外库依赖)
|
||||
|
||||
| 产物 | 路径 |
|
||||
|------|------|
|
||||
| RE 四阶段 | `reverse-engineering/references/re-agent-workflow.md` |
|
||||
| 授权侦察 | `pentest-tools/references/recon-pipeline.md` |
|
||||
| 攻击链门闩 | `attack-chain/references/lifecycle-checklist.md` |
|
||||
| Skill 供应链 | `ops/skill-supply-chain.md` |
|
||||
| 领域覆盖 | `references/domain-coverage-map.md` |
|
||||
|
||||
## 5. 建议优先级(后续迭代)
|
||||
|
||||
| 优先级 | 动作 |
|
||||
|--------|------|
|
||||
| P0 已做 | ops 契约、MASTER 路由、skill 供应链安全文档 |
|
||||
| P1 | 对照 Orizon/ToB 补 pentest 阶段检查单到 attack-chain references |
|
||||
| P2 | 可选「外链 skill 白名单」配置,不进默认路径 |
|
||||
62
reverse-skill-repo/skills/references/domain-coverage-map.md
Normal file
62
reverse-skill-repo/skills/references/domain-coverage-map.md
Normal file
@@ -0,0 +1,62 @@
|
||||
# 本包领域覆盖图(深度优先)
|
||||
|
||||
> 对照社区「几百个微 skill」:我们用 **少量深 skill + 路由 + ops** 覆盖主战场。
|
||||
> 日期:2026-07-18
|
||||
|
||||
## 域 → 本包入口
|
||||
|
||||
| 域 | PRIMARY / 模块 | 备注 |
|
||||
|----|----------------|------|
|
||||
| 移动 Android | `apk-reverse/` `mobile-reverse/` | |
|
||||
| 移动 iOS | `mobile-reverse/` | |
|
||||
| 二进制深挖 | `ida-reverse/` `radare2/` `ghidra-reverse/` | Ghidra = 开源主路径 |
|
||||
| 通用 RE / 反调试 / OLLVM | `reverse-engineering/` | |
|
||||
| .NET | `dotnet-reverse/` | |
|
||||
| 前端 JS / 签名 | `js-reverse/` | |
|
||||
| 浏览器扩展 | `browser-extension-reverse/` | |
|
||||
| DSL/风控 VM | `reverse-engineering/dsl-vm-reverse/` | |
|
||||
| 协议 / PCAP 协议 | `protocol-reverse/` | |
|
||||
| 固件 IoT | `firmware-pentest/` | |
|
||||
| 恶意样本 | `malware-analysis/` | |
|
||||
| 数字取证 / IR | `digital-forensics/` | |
|
||||
| 威胁狩猎 / 蓝队 | `threat-hunting/` | |
|
||||
| 渗透工具 | `pentest-tools/` (+ src-hunter) | |
|
||||
| Windows / AD | `windows-ad/` | |
|
||||
| 云 / 容器 / K8s | `cloud-k8s/` | |
|
||||
| 代码审计 / SAST | `code-audit/` | |
|
||||
| Wi-Fi / 无线 | `wifi-wireless/` | |
|
||||
| OT / ICS | `ot-ics/` | 被动优先;写寄存器默认禁止 |
|
||||
| macOS | `macos-reverse/` | iOS 仍走 mobile-reverse |
|
||||
| 厚客户端 | `thick-client/` | |
|
||||
| Go / Rust 二进制 | `go-rust-reverse/` | |
|
||||
| 硬件调试口 | `hardware-security/` | 交接 firmware-pentest |
|
||||
| 数据库 | `database-security/` | |
|
||||
| 邮件 / 钓鱼 | `email-security/` | |
|
||||
| 联邦身份 SSO | `identity-federation/` | 与 api-security JWT 互补 |
|
||||
| RF / SDR | `radio-sdr/` | 默认只收;非 Wi-Fi |
|
||||
| 多阶段攻击 | `attack-chain/` | |
|
||||
| Pwn | `pwn-chain/` | |
|
||||
| N-day 补丁 | `patch-diff-exploit/` | |
|
||||
| EDR 研究 | `edr-bypass-re/` | |
|
||||
| API | `api-security/` | |
|
||||
| 供应链 SBOM | `supply-chain-security/` | |
|
||||
| LLM/Agent | `llm-security/` | + `ops/skill-supply-chain.md` |
|
||||
| 浏览器自动化 | `browser-automation/` | |
|
||||
| 报告/图 | `docs-generator/` `diagram-generator/` | |
|
||||
| 符号迁移 | `binary-diff/` | |
|
||||
| 作战契约 | `ops/` | **特色** |
|
||||
| CTF 编排 | `CTF-Sandbox-Orchestrator/` | |
|
||||
| 密码学模式识别 | `reverse-engineering` 模式文档 | 与逆向任务共用,不维护独立扩展包 |
|
||||
|
||||
## 明确不整库并入的域(路由未命中时的策略)
|
||||
|
||||
| 域 | 策略 |
|
||||
|----|------|
|
||||
| 纯游戏外挂开发 | 不作为产品方向;Unity 样本仍可走 `reverse-engineering` + seed-014 |
|
||||
| 深度汽车/航空认证级 | 可外链;本包仅有 RF/OT 入口级 |
|
||||
| 纯 GRC/合规长文 | 不替代专业 GRC 工具;可写报告模板引用 |
|
||||
| 800+ ATT&CK 微 skill | 用本表 + ATT&CK 可选标签(Finding 字段) |
|
||||
|
||||
## 与 MITRE ATT&CK(可选)
|
||||
|
||||
Finding 模板允许 `optional_attack: Txxxx`(见 `ops/evidence-finding-path.md`),**不强制**完整 ATT&CK 引擎。
|
||||
Reference in New Issue
Block a user