Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →China standard: L2 Adas Safety. # L2 组合驾驶辅助系统安全要求（强制性国标草案）
.claude/skills/pangzhenying2025-china-l2-adas-safety/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 32% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | -14% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 17% | 0% |
| 属性 | 值 | |------|-----| | 标准编号 | 待定（GB 强标） | | 名称 | 智能网联汽车 组合驾驶辅助系统安全要求 | | 状态 | 项目组草案 v2.1 | | 推荐等级 | P2-推荐入选 | | 性质 | 强制性国标，量产合规核心 | | 核心内容 | L2 ADAS组合功能（ACC+LKA/LCC/TJA/ICA）安全要求 |
适用于具有组合驾驶辅助功能的M类和N类车辆，要求同时具备纵向和横向控制的L2级功能。
组合驾驶辅助功能分类
├── 基础组合：ACC + LKA
├── 进阶组合：ACC + LCC (车道居中)
├── 高级组合：ICA (集成巡航辅助 = ACC + LCC)
├── 拥堵辅助：TJA (交通拥堵辅助)
├── 高速辅助：HWA (高速公路辅助 = ICA + 换道辅助)
└── 相关独立功能：AEB, LDW, BSD, FCTA, RCTAL2强标安全要求架构
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1. ODD定义要求
├── 道路类型/速度范围/天气条件
└── ODD退出策略
2. 系统安全要求
├── 功能安全（基于GB/T 34590）
├── 预期功能安全（基于GB/T 43267）
└── 网络安全
3. DMS强制要求
├── 注意力监测
├── 脱手检测
├── 疲劳检测
└── 防滥用机制
4. HMI要求
├── 系统状态显示
├── 接管提醒分级
└── 功能边界提示
5. 测试验证
├── 封闭场地测试
├── 开放道路测试
└── 仿真测试
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━详细技术内容参见 skills/automotive-china-l2-adas-compliance/china-l2-adas-compliance.md
skills/china-standards/ads-safety/ — ADS强制安全要求skills/china-standards/l2-adas/ — L2 ADAS功能要求（16项）skills/automotive-china-l2-adas-compliance/ — L2合规详细指南| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 9,293 | 4,433 | -52% | 1 | 1 | 0% | 1,747 | 1,497 | -14% | 0 | 0 | — |
case-02 | pass→pass | 14,165 | 13,225 | -7% | 1 | 1 | 0% | 2,486 | 2,916 | +17% | 0 | 0 | — |
case-03 | fail→pass | 6,669 | 5,444 | -18% | 1 | 1 | 0% | 1,217 | 1,601 | +32% | 0 | 0 | — |
case-04 | pass→pass | 5,271 | 2,470 | -53% | 1 | 1 | 0% | 903 | 1,084 | +20% | 0 | 0 | — |
case-05 | pass→pass | 4,522 | 2,447 | -46% | 1 | 1 | 0% | 783 | 1,068 | +36% | 0 | 0 | — |
case-06 | pass→pass | 6,157 | 5,103 | -17% | 1 | 1 | 0% | 1,219 | 1,561 | +28% | 0 | 0 | — |
case-07 | fail→pass | 6,676 | 3,170 | -53% | 1 | 1 | 0% | 1,346 | 1,267 | -6% | 0 | 0 | — |
case-08 | pass→pass | 9,789 | 7,929 | -19% | 1 | 1 | 0% | 1,718 | 1,992 | +16% | 0 | 0 | — |
case-09 | pass→pass | 9,934 | 4,534 | -54% | 1 | 1 | 0% | 1,633 | 1,254 | -23% | 0 | 0 | — |
case-10 | pass→pass | 18,451 | 14,402 | -22% | 1 | 1 | 0% | 3,182 | 3,228 | +1% | 0 | 0 | — |
case-11 | pass→pass | 5,360 | 3,227 | -40% | 1 | 1 | 0% | 920 | 1,080 | +17% | 0 | 0 | — |
case-12 | pass→pass | 7,653 | 8,178 | +7% | 1 | 1 | 0% | 1,284 | 2,030 | +58% | 0 | 0 | — |
case-13 | pass→pass | 12,741 | 14,259 | +12% | 1 | 1 | 0% | 1,900 | 3,021 | +59% | 0 | 0 | — |
case-14 | pass→pass | 9,299 | 7,642 | -18% | 1 | 1 | 0% | 1,679 | 2,010 | +20% | 0 | 0 | — |
case-15 | pass→pass | 10,321 | 3,569 | -65% | 1 | 1 | 0% | 1,761 | 1,261 | -28% | 0 | 0 | — |
case-16 | pass→pass | 6,399 | 4,666 | -27% | 1 | 1 | 0% | 1,201 | 1,338 | +11% | 0 | 0 | — |
case-17 | pass→pass | 8,205 | 5,505 | -33% | 1 | 1 | 0% | 1,341 | 1,503 | +12% | 0 | 0 | — |
case-18 | pass→pass | 6,790 | 2,415 | -64% | 1 | 1 | 0% | 1,235 | 1,087 | -12% | 0 | 0 | — |
case-19 | pass→pass | 17,560 | 19,668 | +12% | 1 | 1 | 0% | 2,969 | 3,748 | +26% | 0 | 0 | — |
case-20 | pass→pass | 7,555 | 9,945 | +32% | 1 | 1 | 0% | 1,312 | 2,122 | +62% | 0 | 0 | — |
case-21 | fail→pass | 7,968 | 8,721 | +9% | 1 | 1 | 0% | 1,384 | 2,069 | +49% | 0 | 0 | — |
case-22 | pass→pass | 4,584 | 3,750 | -18% | 1 | 1 | 0% | 836 | 1,218 | +46% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +14 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.