Skip to content
Back to skills

Semantic Km Creator Cn

ASecurity

Builds evidence-only Kimball semantic layers for enterprise knowledge from APIs, CLIs, or tables/databases with an HTML decision workbench; emits OKF/YAML. Use when 企业知识、知识建模、KM、semantic-km-creator-cn、semantic-creator、semantic layers, dimensional models, ontology, OKF, 语义层、接口转语义、维度建模或本体.

  • 11 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 22, 2026
ai-agentsgogitapidatabase

Works with

  • cli
  • api

Security analysis

A100/100

Pro scans all 8 files and shows the line behind each finding

Scanned September 22, 2026

npx -y skills add ontology-of-everything/SemanticSkills --skill semantic-km-creator-cn --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Semantic Km Creator Cn?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Semantic Km Creator Cn
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ontology-of-everything-semantic-km-creator-cn/badge)](https://www.skillsdirectory.com/skills/ontology-of-everything-semantic-km-creator-cn)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: semantic-km-creator-cn
description: Builds evidence-only Kimball semantic layers for enterprise knowledge from APIs, CLIs, or tables/databases with an HTML decision workbench; emits OKF/YAML. Use when 企业知识、知识建模、KM、semantic-km-creator-cn、semantic-creator、semantic layers, dimensional models, ontology, OKF, 语义层、接口转语义、维度建模或本体.
license: Apache-2.0
metadata:
  language: zh-CN
  translation_of: semantic-km-creator
  version: "0.6.0"
  openclaw:
    homepage: https://github.com/ontology-of-everything/concept-skills/tree/main/skills/cn/semantic-km-creator-cn
---

# Semantic KM Creator(企业知识 · 接口 → 语义层)

把**接口契约**(REST/OpenAPI、CLI、表/DDL)一次通过地建成**受治理语义层**。四个阶段严格串行,每阶段一个参考文件、一组退出条件——满足才前进,不满足不硬闯:

```text
Phase 1 Ingest → Phase 2 Review → Phase 3 Emit → Phase 4 Verify
   ingest.md       review.md      emit-okf.md      verify.md
                                  emit-yaml.md
```

> 唯一真源是接口。**不臆造**字段/粒度/枚举/取值;缺证据停下,一次一问。

## Workflow

### Phase 1 · Ingest(`references/ingest.md`)

接口归一成「操作清单」,每个 op 判定角色(fact / dimension-lookup)。
**退出条件**:清单完整,每个 op 有角色与读写标记。

### Phase 2 · Review(`references/review.md`)

将对象拆为原子决策;显式展示目标、候选、证据、风险、依赖与权衡,并在
事实、维度、度量、路由分区提供固定 YAGNI 判断指导。生成 model JSON 注入
`assets/review-template.html`,由用户确认、选择、修正、补证或拒绝;反馈写入
`amendments.md` 后重出报告。
**退出条件**:收到 `approved:true`,且无未决或阻塞项。出报告后必须停下;
未批准禁止进入 Phase 3。

### Phase 3 · Emit(`references/emit-okf.md` 默认 / `references/emit-yaml.md` 可选)

默认生成 **Google OKF v0.1** bundle;用户明确要求时生成 repo-YAML(Kimball 星型)。字段语义两个目标共享(定义在 `emit-yaml.md` §2–§4)。≥2 个 bundle 时生成薄根索引。
**退出条件**:产物文件完整落盘。

### Phase 4 · Verify(`references/verify.md`)

跑结构检查 + 语义 lint +(如导出 OKF)OKF 硬约束,逐条回报 pass/fail。
**退出条件**:全部 pass;不可机械修复的 fail → 回 Phase 2 One blocking ask,不静默通过。

## Critical Rules

1. **Evidence-only** — 字段/粒度/枚举/取值仅来自接口;缺证据写 `TODO(verify)` 并列入待确认,不编造。
2. **Grain first** — 未锁 grain 不挂维度、不写度量。
3. **One blocking ask** — 仅在改变建模结果处停(一次一问,带候选),不连环问;批量确认走 HTML 评审。
4. **Layer split** — 形态/路由/粒度归语义层;可键入或读取的取值(枚举、code、字段路径、命令)归契约层,语义层只指向。
5. **Stable names** — 事实/维度名确认后不重命名(OKF 概念 ID = 文件路径,改名即断链)。
6. **Decision gate** — 所有决定初始 pending,不得默认通过。Phase 2 出报告后
   停下;仅用户回传 `approved:true` 且无未决/阻塞项时进入 Phase 3。

## Reference Index(按阶段加载)

| 阶段 | 文件 |
| --- | --- |
| Phase 1 物料归一 + 角色判定 | `references/ingest.md` |
| Phase 2 决策建模 + HTML 工作台 + amendments | `references/review.md` |
| Phase 3 默认目标 OKF 布局/映射 | `references/emit-okf.md` |
| Phase 3 可选目标 YAML + 共享字段语义 | `references/emit-yaml.md` |
| Phase 4 全部校验 + 升级回路 | `references/verify.md` |
| 端到端样例 | `references/examples.md` |

Files in this skill

  • SKILL.md3.8 KB
  • assets/review-template.html50.3 KB
  • references/emit-okf.md3.7 KB
  • references/emit-yaml.md3.8 KB
  • references/examples.md3.5 KB
  • references/ingest.md1.4 KB
  • references/review.md8.2 KB
  • references/verify.md1.6 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.