re-ai-triage
Skill152 starsChanged 43 days agoApache-2.0View on GitHub
AI 模型分析第一入口(分流器):识别输入形态——模型文件走文件层逆向(re-ai-model)、 仅有 API 走行为层评估(re-ai-attack)、恶意行为走恶意分析。 触发词:AI模型分析、AI模型安全、模型文件、模型泄露、模型逆向、模型分析入口。
- Installs packages
In this repository123 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- re-address-spaceSkill
- re-ai-attackSkill
- re-ai-modelSkill
- re-analyzeSkill
- re-android-cryptoSkill
- re-android-nativeSkill
- re-angrSkill
- re-anti-analysisSkill
- re-anti-cheatSkill
- re-apkSkill
- re-armSkill
- re-attributionSkill
- re-automotiveSkill
- re-behaviorSkill
- re-binary-coreSkill
- re-binaryninjaSkill
- re-blockchainSkill
- re-browser-extSkill
- re-consoleSkill
- re-cpp-abiSkill
- re-crackingSkill
- re-crash-triageSkill
- re-crypto-decryptSkill
- re-crypto-idSkill
- re-crypto-keysSkill
- re-ctfSkill
- re-deobfuscateSkill
- re-disk-forensicsSkill
- re-doc-malwareSkill
- re-dotnetSkill
- re-drmSkill
- re-ebpfSkill
- re-electronSkill
- re-emulationSkill
- re-evasionSkill
- re-exploitSkill
- re-feedbackSkill
- re-filelessSkill
- re-firmwareSkill
- re-flutterSkill
- re-forensicsSkill
- re-format-elfSkill
- re-format-machoSkill
- re-format-peSkill
- re-fp-runtimeSkill
- re-frida-script-authorSkill
- re-fridaSkill
- re-fuzzingSkill
- re-fw-emulateSkill
- re-fw-extractSkill
- re-fw-rootfsSkill
- re-gameSkill
- re-gdbSkill
- re-ghidraSkill
- re-goSkill
- re-hardware-ioSkill
- re-harmonyosSkill
- re-huntingSkill
- The first 60 shown.
The file
Files
AI 模型分析入口(分流器)
任务分类器(输入形态 → 路径)
拿到 AI 相关目标后先判输入形态,命中即转对应技能:
| 输入形态 | 路径 |
| 拿到模型文件(.onnx / .pt / .pth / .safetensors / .tflite) | → [[re-ai-model]](文件层:格式解析 / 结构还原 / 权重提取 / 文件级水印) |
| 只有 API(无文件,黑盒接口) | → [[re-ai-attack]](行为层:extraction / fingerprint / privacy / robustness 评估;guard 授权前置) |
| 文件 + API 都有 | → 先 [[re-ai-model]](文件侧取证)→ 再 [[re-ai-attack]](行为侧验证) |
| 发现恶意行为(投毒 / 后门 / 恶意载荷 / 下载执行) | → 恶意分析(当前:[[re-ai-model]] 安全边界处理 + [[re-behavior]];未来独立 re-ai-malware 承接) |
| 模型被打包进可执行(PyInstaller/pyarmor 等) | → 先 [[re-binary-core]] 拆包 → 拆出的模型文件回本技能分流 |
何时使用 / 何时不用
- 用:用户说「分析这个模型 / AI 模型安全 / 模型泄露」但未指明输入形态——先分流
- 用:不确定目标是文件层还是行为层时——先识别再转
- 不用:输入形态已明确(直接进 [[re-ai-model]] 或 [[re-ai-attack]],不绕本技能)
- 不用:非 AI 目标(走全局入口 [[re-analyze]])
- 不用:训练 / 微调 / 部署(非逆向)
工具准备
本技能只做识别与分流,工具轻量:
python3 / file —— 模型文件识别
- python3 安装与验证见 [[re-python]] 工具准备
file:Linuxapt install file/dnf install file/pacman -S file(多数预装);验证file --version- 验证:
python3 -c "import struct"(标准库)
模型库探测(判断目标依赖,不加载模型)
pip list 2>/dev/null | grep -iE 'torch|tensorflow|onnx'——确认本机可解析目标格式;缺失时由下游技能引导安装
操作步骤
- 输入形态识别:
- 用户给了路径/文件 → 先
file <path>+xxd <path> | head -2粗看,再转 [[re-ai-model]]。几种形态的可靠信号(都以字节为准,不靠扩展名;细节见 [[re-ai-model]] 步骤 1):- onnx:protobuf 流、无固定魔数,
file报data;首字节是 ir_version 的 tag0x08,第二个字节是版本号、随 ONNX 版本递增(不是常量08);producer_name按规范只 SHOULD 出现 - safetensors:
file报data——前 8 字节是小端头长度,其后才是 JSON 头(xxd -l 16可见长度段与{) - pytorch:新版是 zip 容器,
file报Zip archive(PK\x03\x04头),unzip -l可见data.pkl等条目;老式是裸 pickle 流(无PK头,file常误报XENIX 8086 relocatable or i286 small model)——不直接 load,见 [[re-ai-model]] 坑 2 - tflite:flatbuffers 流,
file报data;字节 4–7 是文件标识符 ASCIITFL3
- onnx:protobuf 流、无固定魔数,
- 用户只有 API 端点/查询能力 → 转 [[re-ai-attack]]
- 两者都有 → 按任务分类器先文件后行为
- 用户给了路径/文件 → 先
- 目标归属确认:文件/API 的持有方与授权(自有 / 授权测试 / CTF·研究)——行为层评估前必须确认([[re-ai-attack]] guard 前置;授权上下文见 triage 第 0 步
RE_AUTH) - 转交与记录:明确转交技能 + 会话变量(
RE_GOAL、RE_AUTH、输入形态标记),不在本技能做深度分析
跨域联合
- [[re-managed]]:本技能是其 AI 分支的第一跳(re-managed → re-ai-triage → re-ai-model / re-ai-attack)
- [[re-ai-model]]:文件层下游
- [[re-ai-attack]]:行为层下游(guard 授权前置)
- [[re-binary-core]]:模型打包进可执行时的拆包前置
- [[re-behavior]]:恶意行为侧协作(投毒/后门的行为判定)
常见坑与陷阱
- 拿文件名猜格式:
.pt可能是 pickle 也可能是 zip 容器(torch 新格式)、.bin可能是任意权重 dump——以file输出与魔数为准,不靠扩展名 - 分流过深:本技能只识别与转交,不展开分析——在分流阶段做深度分析会与下游重复
- 授权前置遗漏:行为层评估([[re-ai-attack]])有 guard 授权要求——转交前先确认
RE_AUTH,未知归属时先询问,不直接进入行为层
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

