Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add pangzhenying2025/hermes-automotive-skills --skill china-ai-safetygit clone --depth 1 https://github.com/pangzhenying2025/hermes-automotive-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/pangzhenying2025/hermes-automotive-skills/china-ai-safety)<a href="https://agentmods.dev/skills/pangzhenying2025/hermes-automotive-skills/china-ai-safety"><img src="https://agentmods.dev/badge/skills/pangzhenying2025/hermes-automotive-skills/china-ai-safety/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/pangzhenying2025/hermes-automotive-skills/china-ai-safety"><img src="https://agentmods.dev/badge/skills/pangzhenying2025/hermes-automotive-skills/china-ai-safety.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00032 | $0.01232 |
| Opus 5 | $0.00016 | $0.00616 |
| Sonnet 5 | $0.00006 | $0.00246 |
| Haiku 4.5 | $0.00003 | $0.00123 |
Grade A, and why
china-ai-safety scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 138 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Ai Safety — China Standard
AI 安全标准集 — ISO PAS 8800 + ISO/IEC TR 5469
标准集一览
| 标准编号 | 名称 | 状态 | 推荐等级 |
|---|---|---|---|
| ISO PAS 8800:2024 | Road Vehicles — Safety and Artificial Intelligence | 已发布 | P1 |
| ISO/IEC TR 5469:2024 | AI — Functional Safety and AI Systems | 已发布 | P1 |
ISO PAS 8800 道路车辆安全与人工智能
核心定位
ISO PAS 8800 是 ISO 26262 在 AI/ML 系统中的适配标准，解决传统功能安全方法无法直接应用于 AI 组件的问题。
核心框架
ISO PAS 8800 AI安全框架
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1. AI组件安全生命周期
├── 需求定义（含AI特有安全需求）
├── 数据管理（训练/验证/测试数据质量）
├── 模型开发（架构选择、训练过程）
├── 验证与确认（AI特有V&V方法）
└── 运行监控（在线监控、OTA更新）
2. AI特有安全考虑
├── 数据质量与偏差
│ ├── 训练数据充分性评估
│ ├── 数据分布偏差检测
│ ├── 标注质量控制
│ └── 域外数据（OOD）处理
├── 模型鲁棒性
│ ├── 对抗样本鲁棒性
│ ├── 分布偏移鲁棒性
│ ├── 传感器退化鲁棒性
│ └── 不确定性量化
├── 可解释性
│ ├── 设计阶段可解释性
│ ├── 运行阶段可解释性
│ └── 事后分析可解释性
└── 性能监控
├── 在线性能指标
├── 异常检测
└── 回退策略
3. 安全论证
├── AI组件安全案例结构
├── 置信度评估
└── 残余风险论证
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
SOFIL（Safety of Functionality Integrity Level）
SOFIL等级 (ISO PAS 8800)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
SOFIL 1E: 最低要求级别
SOFIL 2E: 基础要求级别
SOFIL 3E: 标准要求级别
SOFIL 4E: 高要求级别（L3系统最低要求）
SOFIL 5E: 最高要求级别
SOFIL维度:
- 数据管理严格度
- 模型验证深度
- 运行监控要求
- 安全论证完备性
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
与端到端自动驾驶的关联
ISO PAS 8800 对端到端AD的指导
├── E2E感知模型
│ ├── 训练数据覆盖度评估框架
│ ├── OOD检测机制要求
│ ├── 性能退化监控要求
│ └── 回退至传统算法的切换策略
├── E2E规划模型
│ ├── 行为安全约束（与IEEE 2846关联）
│ ├── 输出可解释性要求
│ └── 安全包络约束
└── 系统级集成
├── AI组件与非AI组件的接口安全
├── 冗余架构要求
└── 运行监控系统设计
ISO/IEC TR 5469 AI与功能安全技术报告
核心内容
ISO/IEC TR 5469 技术报告
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1. AI系统功能安全挑战分析
├── AI的非确定性行为
├── 训练/推理的不一致性
├── 传统安全标准的适用性分析
└── AI系统失效模式分类
2. 安全相关AI系统的生命周期考虑
├── AI安全需求推导
├── AI特有的危害分析方法
├── AI系统的验证与确认策略
└── AI系统的运维安全
3. 与现有安全标准的关系
├── IEC 61508（通用功能安全）
├── ISO 26262（汽车功能安全）
├── ISO 21448（SOTIF）
└── 各行业AI安全标准映射
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 138 lines · 32 tokens per session scan A 76d042302dbd
china-ai-safety is a skill published in the GitHub repository pangzhenying2025/hermes-automotive-skills (5 stars, last pushed 3mo ago), licensed MIT. It adds 32 tokens to every session and 1,232 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
iso26262
ISO 26262 functional-safety expert that operates in two modes: (1) HARA / ASIL determination — enumerate hazardous events from item malfunctions × driving situations, rate Severity (S0–S3), Exposure (E0–E4), Controllability (C0–C3), look up ASIL from ISO 26262-3:2018 Table 4, and produce a HARA report with Safety…
automotive-syseng
When the user wants to analyze automotive requirements, check INCOSE/EARS compliance, review MISRA-C code, assess ADAS levels, or verify ISO 26262/AUTOSAR/SOTIF conformance. Also use when the user says 'check requirements', 'EARS check', 'INCOSE analysis', 'MISRA check', 'ASIL assessment', 'V-model check'…
llamaguard
Meta's 7-8B specialized moderation model for LLM input/output filtering. 6 safety categories - violence/hate, sexual content, weapons, substances, self-harm, criminal planning. 94-95% accuracy. Deploy with vLLM, HuggingFace, Sagemaker. Integrates with NeMo Guardrails.
constitutional-ai
Anthropic's method for training harmless AI through self-improvement. Two-phase approach - supervised learning with self-critique/revision, then RLAIF (RL from AI Feedback). Use for safety alignment, reducing harmful outputs without human labels. Powers Claude's safety system.
implementing-llm-guardrails-for-security
Implements input and output validation guardrails for LLM-powered applications to prevent prompt injection, data leakage, toxic content generation, and hallucinated outputs. Builds a security validation pipeline using NVIDIA NeMo Guardrails Colang definitions, custom Python validators for PII detection and content…
llm-evaluation
Implement comprehensive evaluation strategies for LLM applications using automated metrics, human feedback, and benchmarking. Use when testing LLM performance, measuring AI application quality, or establishing evaluation frameworks.