菜单
Use when designing, reviewing, or implementing interactive charts, diagnostics, evidence plots, or chart-centered question answering in an insurance actuarial workflow.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ryanportfolio/lab-demo --skill designing-actuarial-chart-workspaces命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use when designing, refining, reviewing, or implementing Prediction Lab or insurance actuarial modeling UX, especially AI-agent experiments, evidence, versioning, collaboration, and model review.
Use when comparing, reviewing, reconciling, adjudicating, or finalizing multiple completed actuarial UX synthesis passes, especially Prediction Lab research syntheses, before using findings for product or design decisions.
Use when comparing, reviewing, reconciling, adjudicating, or finalizing multiple completed actuarial UX synthesis passes, especially Prediction Lab research syntheses, before using findings for product or design decisions.
基于 SOC 职业分类
正在显示 SKILL.md
| name | designing-actuarial-chart-workspaces |
| description | Use when designing, reviewing, or implementing interactive charts, diagnostics, evidence plots, or chart-centered question answering in an insurance actuarial workflow. |
| user-invocable | true |
A chart is a working surface, not a report. It must support see → isolate → compare → explain → act → preserve without losing model, data, denominator, or evidence context.
Name the user, decision stage, target, denominator, baseline, slice, guardrail, reviewer, and consequence. Never invent statistics, hierarchy, provenance, or actions; name missing dependencies and disable with a reason.
Every chart must answer:
No question means decoration. Any other gap means incomplete work.
Use one linked selection across plot, table, diagnostics, citations, and review.
| Capability | Rule |
|---|---|
| Hover or focus | Preview value, unit, denominator, and delta. Keep critical facts outside hover. |
| Click, tap, or Enter | Pin selection and drive linked evidence. |
| Range selection | Use for ordered continuous, distribution, geography, or time questions, not unrelated categories. |
| Comparison | Allow only when definitions, populations, periods, and denominators match. |
| Exposure and credibility | Put available weight, sparsity, uncertainty, or instability beside estimates. |
| Guardrails | Overlay only real thresholds governing this decision; show distance and breach direction. Filing and stated goal constraints are real; invented model-quality pass marks (a ΔGini or lift zero-crossing, a closeness cutoff) are not. Render standard error, spread, and exposure context instead, and never color a value pass/fail on a bare point estimate. |
| Drill-down | Require a real hierarchy or supported slice. Preserve parent selection and reset. |
| Ask from selection | Send selection, model, data, filters, run, method, and artifact. Show what the agent read; it cannot approve. |
| Preserve | Carry selection, comparison, filters, table state, provenance, and weak point in the link or review. Label local-only state. Exports carry version identity and reveal when superseded. |
Provide keyboard and touch parity, visible focus, useful targets, color-independent meaning, reduced motion, a semantic exact-values table, and a plain-language summary.
Support empty → loading → ready → refreshing | error | stale. Keep the last valid artifact during refresh. Identify stale or failed context. Never collapse layout or silently mix states.
Verify six questions, applicable interactions, desktop and narrow layouts, keyboard flow, preservation, exact values, all states, and one adversarial case. A skeptical actuary must find the pattern, weakest point, denominator, source, and next action without reconstructing context.
