// the find
limingrui679-design/high-stakes-analytics-decision-lab
A platform-neutral analytical Skill that profiles messy data, selects case-adaptive methods, and produces source-backed visual reports for high-stakes decisions.
An installable Agent Skill plus a Python toolkit that profiles a dataset for grain, keys, quality, and privacy, then selects analysis routes the data can support and writes a report with linked figures and a file trail for rerunning it. It is aimed at analysts and researchers who want high-stakes analysis to stop when the evidence is weak, and the README says plainly that it is a prototype without production validation.
The data gate is the most useful idea here. It has four explicit statuses (ready, ready_with_documented_limitations, needs_user_confirmation, blocked), only safe normalization runs without a named approval, and the processed copy is written beside the untouched source. Reproducibility is more than a claim: make verify rebuilds all fifteen example projects in an isolated copy, figures are tied to claims through chart-map.json, and the 106-test suite covers boundaries most analysis repos skip, such as DNS and SSRF checks and source-parser security. The CI matrix runs Python 3.11 through 3.14 with CodeQL, Bandit, and mypy.
The routing and stopping logic is written as prose rules that an LLM agent is expected to follow. The tests can check the numbers the scripts produce, but nothing here verifies that the agent actually respected the gate. The nested skills/ bundle is generated from root sources, so two copies of the same content have to stay in sync, and the file-count and hash contract adds another thing to keep current. The naming is heavy ('Evidence Intelligence Report', 'Decision Intelligence Brief', 'claim boundary'), and fifteen example projects with 119 figures is a lot to review before you know whether the core tooling fits your own data. The README concedes there is no external adoption or validation, so the 989 stars signal interest more than proven use on real decisions.