skilgen/skilgen

斯基尔根

安装一次Skilgen，在仓库里运行，让它自动保持你的代理技能、文档、企业上下文和批准的MCP连接器的最新状态。

Skilgen 是 Codex、Claude Code、Cursor 及其他编码代理的本地仓库技能系统。它生成并刷新代理在所处理仓库中实际需要的文件。

快速启动

从 PyPI 安装 Skilgen：

如果你想要完整的模型支持运行时，可以导出一个提供者密钥：

初始化仓库，让Skilgen启动其仓库-本地技能系统：

这就是整个理念：

1. 安装斯基尔根
2. 在仓库中运行一次
3. 让 Skilgen 在代码变化时保持仓库-本地代理上下文的新鲜

斯基尔根所做的事

Skilgen 将你的仓库转变为一个代理准备的操作系统，随着代码变化保持最新。

当你在仓库中运行Skilgen时，它可以：

• 生成并刷新 AGENTS.md、FEATURES.md、REPORT.md 和 TRACEABILITY.md
• 生成并刷新项目树 skills/
• 自动检测代码更改和刷新技能
• 用可验证的Skilgen Score来评分技能树的质量。
• 获取企业范围的技能包

Skilgen

Install Skilgen once, run it in your repo, and let it keep your agent skills, docs, enterprise context, and approved MCP connectors up to date automatically.

Skilgen is the repo-local skill system for Codex, Claude Code, Cursor, and other coding agents. It generates and refreshes the files agents actually need inside the repository they are working on.

Quick Start

Install Skilgen from PyPI:

python -m pip install skilgen

Export a provider key if you want the full model-backed runtime:

export OPENAI_API_KEY="your_key_here"
# or
export ANTHROPIC_API_KEY="your_key_here"
# or
export HUGGINGFACEHUB_API_TOKEN="your_token_here"

Initialize the repo and let Skilgen start its repo-local skill system:

skilgen init --project-root .
skilgen deliver --project-root .
skilgen status --project-root .

That’s the whole idea:

1. install Skilgen
2. run it once in a repo
3. let Skilgen keep the repo-local agent context fresh as code changes

What Skilgen Does

Skilgen turns your repository into an agent-ready operating system that stays current as the code changes.

When you run Skilgen in a repo, it can:

• generate and refresh AGENTS.md, FEATURES.md, REPORT.md, and TRACEABILITY.md
• generate and refresh a project skills/ tree
• detect changed code and refresh skills automatically
• score the quality of the skill tree with a verifiable Skilgen Score
• ingest enterprise-wide skill packs
• install and rank external skill ecosystems
• recommend and activate approved official MCP connectors
• decide what an agent should load first before it starts working

Why Teams Use It

Most AI coding sessions fail for the same reasons:

• the agent starts without the real repo context
• guidance gets stale after code changes
• enterprise knowledge is scattered across docs, runbooks, and tribal memory
• tool access is not governed or consistent

Skilgen fixes that by giving the repo a living skill system instead of a one-time prompt.

Skilgen Score

Skilgen Score is the quality bar for a skill tree.

It is a 0-100 score with four 0-25 subscores: