Skip to main content

编译时 Skill 治理工具 — 生成 AGENTS.md,强制执行冲突检测,MCP Server 按需加载

Project description

Skills Orchestrator

编译时 Skill 治理工具 — 管理 AI 编码助手的行为规范,强制执行,防止冲突。

把分散在各处的 .md Skill 文件,编译成 Claude / Cursor / Copilot 能直接读取的 AGENTS.md。配合 MCP Server,让 AI 按需动态加载 Skill,上下文零浪费。


为什么需要它?

问题 没有 Skills Orchestrator 有了之后
Skill 越来越多 手动维护 AGENTS.md,容易遗漏或冲突 build 一条命令自动生成
不同项目用不同规范 到处复制粘贴,版本不同步 Zone 机制,目录自动对应规范
上下文窗口有限 所有 Skill 全量注入,浪费 token MCP Server 按需加载,500 个 Skill 和 5 个消耗相同
两个 Skill 互相冲突 运行时才发现,模型行为不确定 编译时 conflict_with 强制报错

快速开始(5 分钟)

安装

pip install skills-orchestrator

初始化项目

# 在你的项目目录里
cd my-project

# 扫描本地 skills 目录,生成 skills.yaml
skills-orchestrator init

init 会引导你选择 Skill 目录,生成配置文件。

编译生成 AGENTS.md

skills-orchestrator build --config config/skills.yaml
# ✓ 解析完成: 20 skills, 3 zones
# ✓ 使用 Zone: 默认区 (default)
# ✓ 输出: AGENTS.md

把生成的 AGENTS.md 放到项目根目录,Claude / Cursor 会自动读取。

验证配置

skills-orchestrator validate --config config/skills.yaml
# ✓ 配置合法:20 skills,无冲突

MCP Server 集成(Claude Code)

Phase 2 功能:让 Claude 在对话中按需动态加载 Skill,无需全量注入。

启动 MCP Server

skills-orchestrator serve --config config/skills.yaml

配置 Claude Code

在项目的 .claude/settings.json 中添加:

{
  "mcpServers": {
    "skills-orchestrator": {
      "command": "skills-orchestrator",
      "args": ["serve", "--config", "/absolute/path/to/config/skills.yaml"]
    }
  }
}

重启 Claude Code,即可使用以下 4 个工具:

工具 用途
list_skills 查看所有可用 Skill(含 tag 过滤)
search_skills 按需求搜索相关 Skill
get_skill 加载完整 Skill 内容到上下文
suggest_combo 根据任务描述推荐 Skill 组合

测试工具调用

# 不启动 Server,直接在命令行测试
skills-orchestrator mcp-test search_skills "git branch parallel work"
skills-orchestrator mcp-test get_skill karpathy-guidelines
skills-orchestrator mcp-test suggest_combo "部署微服务,需要代码审查和 git 工作流"

Skill 文件格式

每个 .md 文件开头加 YAML frontmatter,Skills Orchestrator 自动发现,无需手动注册:

---
id: my-skill
name: 我的技能
summary: "一句话描述,用于 search 和摘要展示"
tags: [coding, quality]
load_policy: free      # free / require(强制注入)
priority: 80           # 数值越大优先级越高
zones: [default]       # 适用的 Zone
conflict_with: []      # 互斥的 Skill ID 列表
---

# 正文内容(Markdown)
...

目录结构(推荐)

skills/
├── coding/        tdd.md, error-handling.md, api-design.md
├── git/           git-operations.md, git-worktrees.md
├── ops/           deployment-checklist.md, environment-setup.md
├── planning/      brainstorming.md, writing-plans.md
├── quality/       refactoring.md, systematic-debugging.md
└── review/        chinese-code-review.md, security-review.md

配置文件(skills.yaml)

version: "2.0"

# 自动扫描目录,无需手动列出每个 Skill
skill_dirs:
  - ../skills

# Zone:不同目录使用不同规范
zones:
  - id: enterprise
    name: 企业强制区
    load_policy: require       # 该 Zone 内所有 Skill 强制注入
    rules:
      - pattern: "*/internal/*"
      - git_contains: "company.com"

  - id: default
    name: 默认区
    load_policy: free
    rules: []

# 覆盖个别 Skill 的 frontmatter 默认值
overrides: []

# Combo:预定义的 Skill 组合
combos:
  - id: full-dev-workflow
    name: 完整开发工作流
    skills: [brainstorming, writing-plans, git-worktrees, finish-branch]

所有命令

skills-orchestrator build     --config <path>            # 生成 AGENTS.md
skills-orchestrator validate  --config <path>            # 验证,不生成文件
skills-orchestrator status    --config <path>            # 查看 forced/passive/blocked
skills-orchestrator inspect   --workdir <path>           # 检查目录命中哪个 Zone
skills-orchestrator init                                  # 交互式初始化
skills-orchestrator import    <github-url>               # 从 GitHub 导入 Skill
skills-orchestrator serve     --config <path>            # 启动 MCP Server
skills-orchestrator mcp-test  <tool> <args>              # 测试 MCP 工具

从 GitHub 导入 Skill

# 导入单个仓库里的所有 Skill
skills-orchestrator import https://github.com/forrestchang/andrej-karpathy-skills

# 导入后自动放入 skills/external/ 目录

开发

git clone https://github.com/yourname/skills-orchestrator
cd skills-orchestrator
pip install -e ".[dev]"
pytest tests/ -v

路线图

  • 编译时治理(build / validate / zone / conflict)
  • Auto-Discovery from frontmatter
  • MCP Server(list / search / get / suggest_combo)
  • 20 个生产级 Skill 内容库
  • PyPI 正式发布
  • pgvector 语义检索(skill > 50 个时启用)
  • sync --target openclaw 多工具同步

Project details


Release history Release notifications | RSS feed

This version

1.2.0

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

skills_orchestrator-1.2.0.tar.gz (58.2 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

skills_orchestrator-1.2.0-py3-none-any.whl (49.0 kB view details)

Uploaded Python 3

File details

Details for the file skills_orchestrator-1.2.0.tar.gz.

File metadata

  • Download URL: skills_orchestrator-1.2.0.tar.gz
  • Upload date:
  • Size: 58.2 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for skills_orchestrator-1.2.0.tar.gz
Algorithm Hash digest
SHA256 0bed133cf215d376cbf6490d5160097f71b1f1fd01482c01d137e9e478ae211e
MD5 6f3346273f9136b0b748f932cb572aa6
BLAKE2b-256 714be9d007950dd63e3dc311c622642670224c2f8942158003893baed50ebf7b

See more details on using hashes here.

Provenance

The following attestation bundles were made for skills_orchestrator-1.2.0.tar.gz:

Publisher: publish.yml on BambooGap/skills-orchestrator

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file skills_orchestrator-1.2.0-py3-none-any.whl.

File metadata

File hashes

Hashes for skills_orchestrator-1.2.0-py3-none-any.whl
Algorithm Hash digest
SHA256 1672452e7690f9038f05da3e00c5cb89d1a3b19355460bfdf17630245013e1dc
MD5 5b32d31e5349fed018afa06d1d477a5e
BLAKE2b-256 ca03258ba35187edc1957d40d6cd467e666e7df1b0e5cad5f3eac08b9c79666e

See more details on using hashes here.

Provenance

The following attestation bundles were made for skills_orchestrator-1.2.0-py3-none-any.whl:

Publisher: publish.yml on BambooGap/skills-orchestrator

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page