kizemo

Evidence Skill — Data skill for Claude Code

Data community

Evidence BI Core 2026 knowledge, distilled from a real production cutover — 21 chapters: Chinese-schema SQL, nginx 5-patch, postgres_fdw cross-DB, theme.yaml truth, FastMCP self-host, report-page UI c.

How to install Evidence Skill

This entry records only its repository, not the path inside it, so there is no exact command to give. Open kizemo/evidence-skill and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Evidence Skill does

Evidence BI Core 2026 knowledge, distilled from a real production cutover — 21 chapters: Chinese-schema SQL, nginx 5-patch, postgres_fdw cross-DB, theme.yaml truth, FastMCP self-host, report-page UI contracts. LLM accuracy ~22% to ~98%.

Alternatives in Data

  • Clinical Decision Support — Evidence-based clinical decision support tools 6.4k ★
  • L Docs Architecture — Document cross-cutting Lowdefy architecture topics (build pipeline, request lifecycle, etc 3k ★
  • Local News Site — Publish a local news article as a standalone Elementor page, with Google News-grade structured data and image 648 ★

README

evidence-skill — Evidence BI Agent Skill

**English version**: see [README.en.md](README.en.md)

把一个 AI agent 在 **Evidence.dev / Evidence BI Core 2026** 上的工作准确度,从通用 LLM 的 **~22%** 拉到 **生产级 ~98%**(9 类真实任务 A/B 实测,见下文)。 来源: **2026-10-12 production cutover of bi.jury.xin (Sales BI)** + 2026-10-14 R 系列全面调研 + **2026-10-15 全站 docs 镜像(168 页)** + 2026-10-21 报表页 UI 契约(ch21)+ **2026-10-25 token 瘦身(触发成本 -42%)**。

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) [![Skill Version](https://img.shields.io/badge/version-v3.3-blue.svg)](CHANGELOG.md) [![Coverage](https://img.shields.io/badge/docs_mirror-168%20pages-green.svg)](#2026-10-15-docs-mirror-fold-in-168-页) [![Trigger Cost](https://img.shields.io/badge/trigger-2662→1556%20tok-orange.svg)](docs/token-efficiency-audit.md)

**60 秒上手**

npx skills add https://github.com/kizemo/evidence-skill --skill evidence-skill

装到 `.agents/skills/evidence-skill/`,并为 Claude Code 建 symlink(20+ 个 harness 通用)。 装完直接问 agent:`"我的 Evidence 报表页图表不渲染,怎么排查?"` → 它会加载 ch21,给出 4 条「静默不渲染」契约。


目录 / Table of Contents

  1. 项目目标 (Project Goals)
  2. 背景故事 (Why This Skill)
  3. 核心作用 (What It Does)
  4. **对 Agent 使用 Evidence 准确度的提升 (Accuracy Boost)** ← 重点
  5. 关键内容 (Key Content)
  6. 数据规模与覆盖度 (Coverage)
  7. **Token 效率 (Token Efficiency)** ← 2026-10-25
  8. 使用方式 (How to Use)
  9. 安装 (Install)
  10. 文件结构 (File Inventory)
  11. 源数据 (Source)
  12. 维护与版本 (Maintenance)
  13. 贡献 (Contributing)
  14. License

项目目标

**evidence-skill** 是一个专精于 [Evidence.dev](https://docs.evidence.dev) BI 平台的 **agent skill**(可装载到任何支持 agent-skills 协议的 harness 上,如 Claude Code / Codex / Copilot CLI / OpenCode / Amp)。

它要做三件事,目标排序按重要性递减:

  1. 消灭常见错误 — 让 agent 不再重复 LLM 通用知识里的"幻觉",例如:
    • 写 date_substr 而不是 TO_CHAR(date, 'YYYY-MM')
    • 写 kebab-case URL `/components/scat