Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →China standard: Odd. # 自动驾驶系统设计运行条件（ODD）标准
.claude/skills/pangzhenying2025-china-odd/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 30% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 40% | 0% |
| 属性 | 值 | |------|-----| | 标准编号 | 待定（GB/T） | | 名称 | 智能网联汽车 自动驾驶系统设计运行条件 | | 状态 | 工作组征求意见稿 | | 推荐等级 | P2-推荐入选 | | 关联标准 | ISO 34503（ODD Specification），GB/T 40429（分级） |
ODD定义要素（中国标准框架）
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
A. 道路条件
├── 道路类型（高速/城市快速路/城市道路/乡村道路）
├── 道路几何（车道数/车道宽度/曲率/坡度/超高）
├── 道路表面（沥青/混凝土/砂石/状态）
├── 道路设施（标线类型/护栏/中央分隔/路沿）
└── 特殊路段（隧道/桥梁/匝道/交叉口/环岛/收费站）
B. 交通条件
├── 交通参与者类型（机动车/非机动车/行人）
├── 交通流密度（自由流/稳定流/拥堵）
├── 速度范围（最低/最高设计速度）
└── 特殊交通状态（施工区/事故/管制）
C. 环境条件
├── 天气（晴/阴/雨/雪/雾/霾/沙尘/冰雹）
├── 光照（昼间/夜间/黎明/黄昏/隧道）
├── 能见度范围
└── 温度/湿度
D. 数字基础设施
├── 高精地图覆盖与时效性
├── 定位系统精度（GNSS/RTK/惯导）
├── 通信条件（4G/5G/V2X）
└── 云端连接状态
E. 地理区域
├── 经纬度范围
├── 行政区划
└── 特定道路编号
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ODD边界管理策略
├── ODD进入：满足所有ODD条件 → 允许激活
├── ODD内运行：持续监控所有ODD条件
├── ODD退出预警：某项条件接近边界 → 提前提醒
├── ODD退出：条件不满足 → 接管请求/MRM
└── 不确定区域：条件判定不确定 → 保守策略skills/china-standards/ads-safety/ — ADS安全要求skills/china-standards/scenario-safety/ — 场景安全评估| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 26,218 | 25,043 | -4% | 1 | 1 | 0% | 4,532 | 5,025 | +11% | 0 | 0 | — |
case-02 | fail→pass | 25,622 | 28,020 | +9% | 1 | 1 | 0% | 4,077 | 5,286 | +30% | 0 | 0 | — |
case-03 | fail→pass | 16,810 | 13,572 | -19% | 1 | 1 | 0% | 2,731 | 3,054 | +12% | 0 | 0 | — |
case-04 | fail→pass | 18,828 | 4,803 | -74% | 1 | 1 | 0% | 3,490 | 1,580 | -55% | 0 | 0 | — |
case-05 | pass→pass | 9,915 | 5,180 | -48% | 1 | 1 | 0% | 1,762 | 1,565 | -11% | 0 | 0 | — |
case-06 | fail→pass | 12,050 | 4,947 | -59% | 1 | 1 | 0% | 1,978 | 1,483 | -25% | 0 | 0 | — |
case-07 | pass→pass | 9,691 | 5,200 | -46% | 1 | 1 | 0% | 1,500 | 1,556 | +4% | 0 | 0 | — |
case-08 | pass→pass | 19,976 | 18,630 | -7% | 1 | 1 | 0% | 3,077 | 3,530 | +15% | 0 | 0 | — |
case-09 | pass→pass | 16,273 | 14,392 | -12% | 1 | 1 | 0% | 2,549 | 2,940 | +15% | 0 | 0 | — |
case-10 | pass→pass | 8,093 | 2,105 | -74% | 1 | 1 | 0% | 1,363 | 963 | -29% | 0 | 0 | — |
case-11 | pass→pass | 4,515 | 4,072 | -10% | 1 | 1 | 0% | 765 | 1,228 | +61% | 0 | 0 | — |
case-12 | pass→pass | 14,641 | 12,092 | -17% | 1 | 1 | 0% | 2,356 | 2,571 | +9% | 0 | 0 | — |
case-13 | pass→pass | 14,085 | 4,995 | -65% | 1 | 1 | 0% | 2,466 | 1,506 | -39% | 0 | 0 | — |
case-14 | pass→pass | 11,082 | 3,176 | -71% | 1 | 1 | 0% | 1,773 | 1,117 | -37% | 0 | 0 | — |
case-15 | pass→pass | 18,406 | 15,749 | -14% | 1 | 1 | 0% | 3,069 | 3,283 | +7% | 0 | 0 | — |
case-16 | pass→pass | 15,265 | 13,958 | -9% | 1 | 1 | 0% | 2,522 | 2,857 | +13% | 0 | 0 | — |
case-17 | fail→pass | 16,238 | 19,515 | +20% | 1 | 1 | 0% | 2,556 | 3,589 | +40% | 0 | 0 | — |
case-18 | pass→pass | 10,801 | 8,193 | -24% | 1 | 1 | 0% | 1,790 | 1,872 | +5% | 0 | 0 | — |
case-19 | fail→pass | 8,118 | 1,510 | -81% | 1 | 1 | 0% | 1,384 | 845 | -39% | 0 | 0 | — |
case-20 | pass→pass | 25,380 | 31,261 | +23% | 1 | 1 | 0% | 4,352 | 5,738 | +32% | 0 | 0 | — |
case-21 | pass→pass | 22,138 | 21,712 | -2% | 1 | 1 | 0% | 3,958 | 4,769 | +20% | 0 | 0 | — |
case-22 | pass→pass | 27,758 | 26,791 | -3% | 1 | 1 | 0% | 4,733 | 5,094 | +8% | 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 +27 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.