# Screening

> 进行单变量筛选并形成候选解释变量清单（feature selection 特征筛选）

- Skill: `zafer-liu/screening` (Agent Skill)
- Install (CLI): `npx skillmds add zafer-liu/screening`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zafer-liu/screening/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: Zafer-Liu (https://skillmd.com/u/zafer-liu)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/zafer-liu/screening

---

# 单变量筛选

确认目标变量与候选字段，对每个候选变量执行适当的单变量检验或回归。报告效应方向、效应量、显著性、缺失率和样本量；多重比较时提示假阳性风险，不直接宣称因果。

## Tool routing

1. Use `get_schema` to identify the target variable, candidate predictors, and source table.
2. Use `query_data` to verify field names, missingness, and candidate variable types.
3. Use `run_analysis` with `analysis_name="Univariate_Screening"` for the screening computation.
4. Use `generate_chart` on screening rankings or effect result tables after `run_analysis` succeeds.

## Implementation reference

- Tool entry: `agent/tools/business/data.py::_tool_run_analysis`
- Analysis registry: `Function/Analyze/registry.py`
- Analysis implementation: `Function/Analyze/Univariate_Screening/analyze.py`
- Chart implementation: `Function/Charts_generation/chart_generate.py`