aedelon/claude-code-blueprint 是从 https://github.com/aedelon/claude-code-blueprint. 导入的特工技能候选
概览
aedelon/claude-code-blueprint 是从 https://github.com/aedelon/claude-code-blueprint. 导入的特工技能候选
概述（AI 翻译）
一个生产级 Claude Code 配置：技能、代理、钩子、规则和权限作为一个系统工作，而非孤立的功能。
文章配套仓库：“我把Claude代码变成了一个操作系统。这是蓝图。”*
建筑
每一层相互强化。反幻觉协议之所以有效，因为Context7是一个可访问的MCP服务器。安全钩子之所以有效，因为静态权限定义了边界。多智能体头脑风暴之所以有效，是因为智能体拥有专用工具和模型。
包含内容
| 组件 | 计数 | 描述 |
|---|---|---|
| CLAUDE.md | 1 | 内核。反幻觉协议、置信等级、安全规则、代码标准。 |
| 规则 | 2 | 路径范围约定：python.md 和 typescript.md。仅在编辑匹配文件时加载。 |
| 技能 | 8 / 32 | 代表性子集。每个都遵循五相AAPEV模式。 |
| 代理 | 4 | 跨不同领域的安全示例。 |
| 钩子 | 11 | 9 完整 + 2 净化（安全模式通用化）。 |
| 命令 | 3 | 斩击命令：/agent、/docs、/review。 |
本页由 AI 翻译，技术细节请以原文为准。
README
以下为原文 README 节选，完整内容请查看原文。
Claude Code Blueprint
A production-grade Claude Code configuration: skills, agents, hooks, rules, and permissions working as a system, not as isolated features.
Companion repo for the article: “I Turned Claude Code Into an Operating System. Here’s the Blueprint.”
Architecture
+---------------------------------------------------+
| LAYER 6: MCP |
| 6 servers (context7, fetch, git, |
| huggingface, jupyter, ide) |
+---------------------------------------------------+
| LAYER 5: SECURITY |
| 40 allow + 38 deny + hooks (17 events) |
| + regex secrets + fail-closed trap |
+---------------------------------------------------+
| LAYER 4: AGENTS |
| Specialized sub-agents (by domain) |
| Multi-agent brainstorm (parallel) |
+---------------------------------------------------+
| LAYER 3: SKILLS |
| 32 skills (10 categories) auto-loaded |
| Common pattern: 5 phases AAPEV |
+---------------------------------------------------+
| LAYER 2: MEMORY |
| memory/ persistent + rules/ per project |
| Cross-session learning |
+---------------------------------------------------+
| LAYER 1: CLAUDE.md (KERNEL) |
| Anti-hallucination + Confidence levels + |
| Toolchain + Code standards + /compact rules |
+---------------------------------------------------+
Each layer reinforces the others. The anti-hallucination protocol works because Context7 is an accessible MCP server. Security hooks work because static permissions define a perimeter. Multi-agent brainstorm works because agents have dedicated tools and models.
What’s Included
| Component | Count | Description |
|---|---|---|
| CLAUDE.md | 1 | The kernel. Anti-hallucination protocol, confidence levels, security rules, code standards. |
| Rules | 2 | Path-scoped conventions: python.md and typescript.md. Load only when editing matching files. |
| Skills | 8 / 32 | Representative subset. Each follows the 5-phase AAPEV pattern. |
| Agents | 4 | Safe examples across different domains. |
| Hooks | 11 | 9 complete + 2 sanitized (security patterns generalized). |
| Commands | 3 | Slash commands: /agent, /docs, /review. |
| Settings | 1 | Template with permission structure (adapt to your needs). |
| Memory | dir | Empty directory with .gitkeep. Claude writes here across sessions. |
What’s NOT Included
推荐工具
换一个关键词，或者移除筛选条件。
安装
npx skillfish add aedelon/claude-code-blueprint