Skip to content
Back to skills

Concept Guardrails Cn

ASecurity

Run concept-spec guardrails (audit, concept, drift, pipeline, sync, map) only when the user explicitly invokes $concept-guardrails-cn. / 仅在用户显式调用 $concept-guardrails-cn 时运行概念规格护栏。

  • 11 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 22, 2026
ai-agentsrailsgit

Works with

  • claude code

Security analysis

A100/100

Pro scans all 16 files and shows the line behind each finding

Scanned September 24, 2026

npx -y skills add ontology-of-everything/SemanticSkills --skill concept-guardrails-cn --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Concept Guardrails Cn?

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

Security grade badge for Concept Guardrails Cn
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ontology-of-everything-concept-guardrails-cn/badge)](https://www.skillsdirectory.com/skills/ontology-of-everything-concept-guardrails-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: concept-guardrails-cn
description: Run concept-spec guardrails (audit, concept, drift, pipeline, sync, map) only when the user explicitly invokes $concept-guardrails-cn. / 仅在用户显式调用 $concept-guardrails-cn 时运行概念规格护栏。
compatibility: 只需读写文件与 Grep/Glob;可选的边界注入运行时需要 Claude Code hooks 与 jq。
metadata:
  language: zh-CN
  translation_of: concept-guardrails
  author: ontology-of-everything
  version: "0.32.1"
  openclaw:
    homepage: https://github.com/ontology-of-everything/concept-skills/tree/main/skills/cn/concept-guardrails-cn
---

# 概念护栏

焦点问题:现有规格是否清楚表达模块契约,并与事实及派生地图对应?仅在显式调用 `$concept-guardrails-cn` 时运行。流程改编自 [wyx v0.26.0](https://github.com/jlifyio/wyx)(MIT,见 `LICENSE.upstream`);规格用本仓 Jackson 方言,runtime 保留上游脚本。

## 规格权威

- **CONCEPT.md**:purpose/state/actions/OP 定义单一目的的行为单元;notes 记录应用背景。
- **SYNCS.md**:app/include/sync 记录应用协调,when/where/then 定义规则,`// flow:` 分组。
- **PIPELINE.md**:记录来源、阶段、输出、不变量、触发与 data boundary;不替代 sync。
- **ARCHITECTURE.md**:从规格派生导航;产品依赖来自总体 PRD,不从同步边推导。
- **回填/漂移**:前者从实现重建可证实契约,后者对比契约与实现;都不证明目的适切性。

**同步可收窄行为,不可扩展契约**:回填须标出违约协调,不能把它转录为合法规则。规格覆盖不证明设计正确,地图准确不证明实现符合规格;推定目的标为推断,不能用理想 OP 合理化缺陷。

## 模式与场景

`$concept-guardrails-cn <模式> [路径或描述]`;路径 → 回填,描述 → 新建,无参数 → 发现候选。未指定模式或项目无规格时先 audit。

| 模式 | 产出 / 必读参考 |
| --- | --- |
| `audit` | 只读覆盖与命令计划:[audit.md](references/audit.md) |
| `concept` | 概念候选或 CONCEPT:[concept.md](references/concept.md) |
| `drift` | 只读漂移报告:[drift-detection.md](references/drift-detection.md) |
| `pipeline` | 管道候选或 PIPELINE:[pipeline.md](references/pipeline.md) |
| `sync` | 协调候选或 SYNCS:[sync.md](references/sync.md) |
| `map` | 派生地图:[map.md](references/map.md) |

复杂建模 → concept-design-cn;确认模型批量转录 → concept-prd-cn;目的适切性、全面独立性与组合审计 → concept-audit-cn。伴生技能不可用时报告交接内容。

## 执行约束

- 先读规格、实现和项目约定。回填标出现状缺陷;已确认的新行为先改规格再改实现。
- 写入前呈现草案/diff;已有创建/修复授权直接完成,只有未决模型取舍或超出授权才确认。发现模式只列候选;只读审计不写历史或修复。
- 每概念一份权威规格;所属 PIPELINE 同目录,跨概念管道放组合层;每个 sync 包一份 SYNCS,按 flow 归组。规格变化时提示已有地图需重建,已授权则完成。
- 生成/读取 CONCEPT、SYNCS 或核对管道接口前,读 [规格契约](references/spec-format.md),统一核对行为、接口与方言兼容。
- wyx 旧 interactions/dependencies/known coupling/dispatching/coordination graph/sync: 段标待迁移,不混读;授权迁移时一起重写 CONCEPT/SYNCS/PIPELINE,保留触发、绑定、效果与失败语义。

## 完成条件

产出符合所选模式,定义、关系和引用有据可查;回填区分事实、推断与未决。报告扫描范围和修复方向;地图依内容重建,不以时间戳证明新鲜度。

启用自动注入时读 [hooks-runtime.md](references/hooks-runtime.md):hooks 仅列规格、提示陈旧及注入 PIPELINE data boundary,不验证语义边界。

Files in this skill

  • LICENSE.upstream1 KB
  • SKILL.md4.2 KB
  • agents/openai.yaml389 B
  • references/audit.md1.9 KB
  • references/concept.md2.2 KB
  • references/drift-detection.md5.8 KB
  • references/hooks-runtime.md2.1 KB
  • references/map.md2.8 KB
  • references/pipeline.md2.2 KB
  • references/spec-format.md7.7 KB
  • references/sync.md2.6 KB
  • runtime/.claude-plugin/plugin.json539 B
  • runtime/hooks/hooks.json970 B
  • runtime/scripts/drift-context.sh10 KB
  • runtime/scripts/post-check.sh4.4 KB
  • runtime/scripts/session-start.sh14.1 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.