Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →日本語翻訳：このファイルは product-lens 用の日本語翻訳が必要です
.claude/skills/affaan-m-product-lens/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-18 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 6% | 0% |
此通道负责产品诊断，而非编写可实施的规格文档。
若用户需要持久的 PRD 到 SRS 或能力契约文档，请移交至 product-capability。
类似 YC 办公时间但自动化。提出尖锐问题：
1. 这是为谁准备的？（具体的人，而非“开发者”）
2. 痛点是什么？（量化：频率、严重程度、当前应对方式？）
3. 为什么是现在？（什么变化使其成为可能/必要？）
4. 10星版本是什么？（如果资金/时间无限）
5. MVP是什么？（能验证假设的最小方案）
6. 反目标是什么？（明确不构建什么？）
7. 如何判断有效？（指标，而非感觉）输出：一份包含答案、风险及"可行/不可行"建议的 PRODUCT-BRIEF.md。
若结果为"是，构建此功能"，下一通道为 product-capability，而非更多创始人表演。
以创始人视角审视当前项目：
1. 阅读 README、CLAUDE.md、package.json、最近的提交
2. 推断：这个项目试图成为什么？
3. 评分：产品市场契合度信号（0-10分）
- 使用增长轨迹
- 留存指标（重复贡献者、回访用户）
- 收入信号（定价页面、计费代码、Stripe集成）
- 竞争护城河（什么难以复制？）
4. 识别：能让这个项目实现10倍增长的关键因素
5. 标记：你正在构建但无关紧要的内容映射实际用户体验：
1. 以新用户身份克隆/安装产品
2. 记录每一个摩擦点（令人困惑的步骤、错误、缺失的文档）
3. 为每个步骤计时
4. 与竞争对手的入门流程进行比较
5. 评分：价值实现时间（用户需要多久才能获得首次成功？）
6. 建议：入门流程的三大修复方案当你有 10 个想法却需选出 2 个时：
1. 列出所有候选功能
2. 对每个功能进行评分：影响（1-5）× 信心（1-5）÷ 工作量（1-5）
3. 按 ICE 分数排序
4. 应用约束条件：时间窗口、团队规模、依赖关系
5. 输出：带有理由的优先级路线图所有模式均输出可操作文档，而非长篇大论。每条建议均附带具体下一步行动。
配合使用：
/browser-qa 验证用户旅程审计结果/design-system audit 进行视觉优化评估/canary-watch 用于发布后监控product-capability 当产品简报需转化为可实施的能力计划时| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→pass | 19,043 | 16,044 | -16% | 1 | 1 | 0% | 2,836 | 3,146 | +11% | 0 | 0 | — |
case-17 | pass→pass | 16,867 | 18,207 | +8% | 1 | 1 | 0% | 2,468 | 3,344 | +35% | 0 | 0 | — |
case-01 | fail→pass | 22,932 | 17,655 | -23% | 1 | 1 | 0% | 3,473 | 3,505 | +1% | 0 | 0 | — |
case-02 | fail→fail | 27,399 | 12,252 | -55% | 1 | 1 | 0% | 5,007 | 2,509 | -50% | 0 | 0 | — |
case-03 | fail→fail | 22,511 | 9,539 | -58% | 1 | 1 | 0% | 3,169 | 2,129 | -33% | 0 | 0 | — |
case-04 | fail→pass | 22,660 | 10,673 | -53% | 1 | 1 | 0% | 2,961 | 2,480 | -16% | 0 | 0 | — |
case-05 | fail→pass | 31,883 | 14,442 | -55% | 1 | 1 | 0% | 5,133 | 2,783 | -46% | 0 | 0 | — |
case-06 | fail→pass | 27,238 | 24,406 | -10% | 1 | 1 | 0% | 4,401 | 4,673 | +6% | 0 | 0 | — |
case-07 | fail→pass | 26,278 | 9,788 | -63% | 1 | 1 | 0% | 5,130 | 2,275 | -56% | 0 | 0 | — |
case-08 | pass→pass | 23,251 | 21,397 | -8% | 1 | 1 | 0% | 3,044 | 3,697 | +21% | 0 | 0 | — |
case-09 | pass→pass | 22,686 | 18,007 | -21% | 1 | 1 | 0% | 3,001 | 3,086 | +3% | 0 | 0 | — |
case-10 | pass→pass | 21,008 | 19,767 | -6% | 1 | 1 | 0% | 2,826 | 3,342 | +18% | 0 | 0 | — |
case-11 | pass→pass | 18,021 | 16,759 | -7% | 1 | 1 | 0% | 2,590 | 3,123 | +21% | 0 | 0 | — |
case-12 | pass→pass | 23,578 | 22,089 | -6% | 1 | 1 | 0% | 3,236 | 3,839 | +19% | 0 | 0 | — |
case-13 | pass→pass | 17,641 | 16,415 | -7% | 1 | 1 | 0% | 2,525 | 3,233 | +28% | 0 | 0 | — |
case-14 | fail→fail | 22,585 | 17,386 | -23% | 1 | 1 | 0% | 3,400 | 3,455 | +2% | 0 | 0 | — |
case-15 | fail→fail | 19,912 | 17,089 | -14% | 1 | 1 | 0% | 2,835 | 3,309 | +17% | 0 | 0 | — |
case-16 | fail→fail | 17,518 | 17,178 | -2% | 1 | 1 | 0% | 2,879 | 3,361 | +17% | 0 | 0 | — |
case-19 | fail→pass | 19,250 | 10,419 | -46% | 1 | 1 | 0% | 2,932 | 2,448 | -17% | 0 | 0 | — |
case-20 | fail→pass | 15,957 | 14,221 | -11% | 1 | 1 | 0% | 2,384 | 3,062 | +28% | 0 | 0 | — |
case-21 | pass→pass | 19,649 | 17,363 | -12% | 1 | 1 | 0% | 2,726 | 3,165 | +16% | 0 | 0 | — |
case-22 | pass→pass | 16,734 | 13,853 | -17% | 1 | 1 | 0% | 2,560 | 2,948 | +15% | 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 +36 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.