Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →从二维光路示意图与厂家 CAD 构建、审核和修订具有物理可信度的 Blender 光学平台。适用于 Thorlabs 兼容镜架、透镜架、分束器、支柱、光纤耦合、孔径、探测器、完整光路拓扑、fail-closed 证据和 Nature 风格光子学渲染。
.claude/skills/k-telux-thorlabs-blender-optical-path-zh/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 121% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 66% | 0% |
把测量需求或二维示意图转换成可解释、可独立审计的 Blender 光学平台。
英文版是技术权威源。涉及几何时读取 ../../skills/thorlabs-blender-optical-path/references/physical-gates.md；验收前读取 evidence-contract.md；修订旧场景时读取 history-derived-rules.md；完整案例见 project-case-study.md；整机运行前读取 end-to-end-workflow.md；多轮规模/发布资格测试读取 multi-run-qualification.md。
UNVERIFIED 或 BLOCKED。文件存在、CAD 导入、进程成功、AABB 接触和自报文本都不是证据。先区分新测量设计、原图重建、既有场景修正与纯渲染。用户要求“参考旧示例的精度生成另一种光路”时，旧示例只作画质比较，需从空场景建立新拓扑、生成器与资产映射。optics-only 包含光学探测器和机械支撑，排除电路及电气/数据信号可视化。详细要求见新设计与渲染。
scripts/workflow_ledger.py 初始化一个整机 run spec、状态账本和 append-only 事件哈希链。schematic node -> 实验角色 -> 真实资产 -> 光/光纤/电端口 -> 支撑路径。BLOCKED，不得异常退出或静默折叠重复项。active_states 或哈希化确定型展开，每个 state 都要有明确 ray template。光束、柔性光纤和电缆必须是不同对象族，仅生成请求范围内的类别。孔径必须真实开放。删除重复 surrogate 和无角色 placeholder。低成本诊断预览可用于检查器件精度、可见光路与构图，但必须注明尚未闭合的物理证据；最终发表级渲染在物理门通过后完成。保留材质槽与多边形索引语义，逐器件近景和逐分支检查；全图清晰度、边缘密度、颜色像素数不能单独证明建模精度或光束连续。实际 2K 图不能满足 4K 交付要求。
达到用户要求与硬门槛后冻结一次交付并发布；只在输入、依赖、runtime 变化或检查失败时复验受影响部分。Skill 文档更新可以记录未通过模型的经验，不能转移模型发布信用。案例数值阈值不得直接成为通用标准。
状态：PASS 表示所有适用规则有新鲜证据；PARTIAL/SCOPED 只代表局部，必须列出 blocker 且 final/release 保持 false；UNVERIFIED 表示证据不足；BLOCKED 表示已知失败。禁止把脱敏通过、局部通过或 CAD 导入成功写成整机最终通过。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 42,599 | 45,954 | +8% | 1 | 1 | 0% | 7,343 | 9,069 | +24% | 0 | 0 | — |
case-02 | fail→fail | 81,271 | 46,530 | -43% | 1 | 1 | 0% | 2,376 | 9,023 | +280% | 0 | 0 | — |
case-03 | fail→fail | 49,269 | 52,923 | +7% | 1 | 1 | 0% | 8,273 | 6,942 | -16% | 0 | 0 | — |
case-04 | fail→pass | 24,213 | 40,395 | +67% | 1 | 1 | 0% | 3,406 | 7,527 | +121% | 0 | 0 | — |
case-05 | fail→fail | 28,620 | 39,158 | +37% | 1 | 1 | 0% | 4,785 | 7,509 | +57% | 0 | 0 | — |
case-06 | fail→fail | 43,198 | 32,164 | -26% | 1 | 1 | 0% | 1,916 | 6,783 | +254% | 0 | 0 | — |
case-07 | fail→pass | 19,094 | 13,742 | -28% | 1 | 1 | 0% | 2,555 | 3,212 | +26% | 0 | 0 | — |
case-08 | fail→pass | 15,137 | 10,532 | -30% | 1 | 1 | 0% | 2,138 | 2,473 | +16% | 0 | 0 | — |
case-09 | fail→pass | 22,795 | 17,808 | -22% | 1 | 1 | 0% | 2,986 | 3,417 | +14% | 0 | 0 | — |
case-10 | fail→pass | 19,459 | 21,907 | +13% | 1 | 1 | 0% | 2,687 | 4,450 | +66% | 0 | 0 | — |
case-11 | fail→pass | 18,701 | 21,601 | +16% | 1 | 1 | 0% | 2,708 | 4,476 | +65% | 0 | 0 | — |
case-12 | pass→pass | 16,704 | 15,277 | -9% | 1 | 1 | 0% | 2,359 | 3,453 | +46% | 0 | 0 | — |
case-13 | fail→pass | 20,950 | 22,333 | +7% | 1 | 1 | 0% | 2,794 | 4,133 | +48% | 0 | 0 | — |
case-14 | fail→pass | 24,064 | 23,425 | -3% | 1 | 1 | 0% | 3,147 | 4,595 | +46% | 0 | 0 | — |
case-15 | fail→pass | 14,106 | 12,306 | -13% | 1 | 1 | 0% | 2,096 | 2,974 | +42% | 0 | 0 | — |
case-16 | fail→pass | 21,197 | 10,979 | -48% | 1 | 1 | 0% | 2,621 | 2,767 | +6% | 0 | 0 | — |
case-17 | pass→pass | 15,916 | 12,772 | -20% | 1 | 1 | 0% | 2,294 | 3,162 | +38% | 0 | 0 | — |
case-18 | fail→pass | 20,720 | 14,442 | -30% | 1 | 1 | 0% | 2,640 | 3,153 | +19% | 0 | 0 | — |
case-19 | fail→pass | 18,171 | 16,373 | -10% | 1 | 1 | 0% | 2,221 | 3,354 | +51% | 0 | 0 | — |
case-20 | pass→pass | 16,843 | 9,905 | -41% | 1 | 1 | 0% | 2,057 | 2,600 | +26% | 0 | 0 | — |
case-21 | pass→pass | 13,167 | 13,565 | +3% | 1 | 1 | 0% | 1,854 | 3,081 | +66% | 0 | 0 | — |
case-22 | pass→pass | 20,009 | 15,898 | -21% | 1 | 1 | 0% | 2,574 | 3,356 | +30% | 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, and 20 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 20 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/28/2026 | +32% |
| gemini-3.6-flash | verified | 8/22/2026 | +18% |
Other measured skills in the registry, with their headline benchmark lift.