
Skill Shelf (Rust singleton daemon)
io.github.halflifezyf2680Published by GitHub account halflifezyf2680v2.0.5Updated Oct 28, 2026
Progressive group routing for 524 professional skills via Rust daemon stdio shim.
The publisher's own server: rust.wiki
Overview
Local skill library for AI agents: a Rust daemon exposes 524 professional skills through 7 MCP tools, keeping context small by loading skills on demand.
- What you can do
- Browse the shelf: list skill groups and the skills inside a group, with status info
- Search skills by keyword across the whole library, with fallback language retry and CJK handling
- Read a skill's summary or its full text, references and assets
- Install skill packages from a local directory and validate their integrity, optionally cleaning broken ones
- Create, update or delete storage groups and move skills between them
- When to use it
- Useful when an assistant should draw on a large, curated set of professional skills or a team's own accumulated know-how without paying the context cost of hundreds of always-loaded descriptions. Also fits setups where several MCP clients should share one skill index and one background process.
- Requirements
- Runs locally over stdio as a Node.js shim that starts a bundled Rust daemon; release packages include the platform binary, so no Rust toolchain is needed. Optional environment variables: SKILL_SHELF_ROOT, SKILL_SHELF_SEARCH_LIMIT, SKILL_SHELF_MAX_RELATED_SKILLS, SKILL_SHELF_WATCH. Desktop MCP clients only.
Installation
In SourceWeft
- Open Skill Shelf (Rust singleton daemon) in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
Skill Shelf
上千个专业技能,只占 context 7 个工具定义。
给 AI agent 装 skill,最痛的问题是:skill 越多,context 越胖。每个 skill 的 description 常驻上下文,几百个 skill 就是几万 tokens 白白浪费,每次对话都背着跑。开多个客户端还各跑各的进程,内存翻倍。更烦的是,装了一堆 skill 自己都记不住哪个是干嘛的、哪些真的好用哪些是花架子,最后还是全靠人去翻文档。
Skill Shelf 连这个问题一起解决了:LLM 看见 MCP 工具就会自己去查、自己去用。你不需要记住每个 skill 的内容,也不需要判断什么时候该用——LLM 遇到合适的场景自己会去搜索和加载。
Skill Shelf 的解法:skill 全部存本地仓库,context 里只有 7 个工具定义。需要时搜索加载,不需要时零开销。Rust 单例 daemon,一个进程服务所有 MCP 客户端——Claude Code、Codex、Cursor、Windsurf 同时开也只有一个后台进程。
不只是用内置的几百个 skill 和 group。工作中积累的经验、踩过的坑、反复用的工作流,都可以整理成 skill 入库——一份 Markdown 文件就是一个 skill。内置的 18 个组不够用就自己建,manage_group 创建自定义分组,install_skills 批量入库。把自己团队的 know-how 变成可复用的 skill 库。
安装
从 GitHub Release 使用
下载与你的平台匹配的 release 包并解压:
MCP 客户端配置:
release 包已经内置对应平台的 Rust daemon 二进制,不需要安装 Rust 或执行 cargo build。
从源码使用
MCP 客户端配置:
也可以先 npm link,再使用 npm bin:
支持 Claude Code(~/.claude.json)、Claude Desktop(claude_desktop_config.json)、Cursor、Windsurf 等所有 MCP 兼容客户端。每个客户端各自启动一个 stdio shim,共享同一个 daemon 进程。
首次调用时 stdio shim 会自动拉起 Rust daemon。多个 MCP 客户端会共享同一个 daemon,不需要分别管理后台进程。
使用
配置完成后,在 MCP 客户端里先调用:
正常结果应包含:
常用流程:
如果不知道该进哪个组,直接搜索:
安装自己的 skill 包:
如果你希望把 skill 库放到包目录以外的位置,设置 SKILL_SHELF_ROOT。例如:
架构
- 单进程: 整个仓库只有一个 Rust daemon 在跑,不会因为开多个客户端就跑出几十个 Node 进程
- stdio shim: 每个 MCP 客户端启动一个极轻量的 shim 进程(只做 stdin↔IPC 转发),真正的业务逻辑全在 daemon 里
- 共享状态: 所有客户端共享同一个 skill 索引和缓存,热重载一次全局生效
- workspace 隔离: 不同
SKILL_SHELF_ROOT的配置各自独立,互不干扰
路由协议
先看组,再看组内 skill,最后才用 search_skills 兜底。
语言策略
搜索会先用用户语言尝试,没结果时再换英文重试。skill 作者无需为每个 skill 写多语言 description。
中文搜索
search_skills 支持两种输入方式:
- 空格分词(推荐):
品牌 视觉 设计 - 连续输入(兜底):
品牌设计视觉— 自动切分为 CJK bigram,匹配精度略低于手动分词
工具清单(7 个)
只读
写操作
组体系
18 个内置组:
engineering · design · product · project-management · marketing · paid-media · sales · finance · legal-compliance · hr-talent · support-operations · supply-chain · academic-research · testing-qa · spatial-gaming · specialized-domain · game-studios · creative-media
安装 skill 时,如果 SKILL.md frontmatter 未指定 group,工具会返回 skill 描述和可用组列表,由 LLM 选择最合适的组。
存储结构
Skill 包格式
每个 skill 是一个包含 SKILL.md 的目录:
name 和 description 是必填 frontmatter 字段。group 可选,不填时由 LLM 在安装时分类。
热重载
daemon 启动时自动监听 packages/ 目录变更,新增、修改、删除 skill 后索引自动更新,无需重启。
CLI 命令
环境变量
致谢
部分 Skill 内容来源于以下开源项目:
- agency-agents-zh(MIT License)— 211 个中文 AI 专家智能体
- awesome-design-md(MIT License)— 品牌设计系统 markdown 文件
- scientific-agent-skills(MIT License)— 139 个科学研究技能(生物信息、药物发现、量子计算等)
Source: README.md at commit 36f60a5
Tools
0Version history
1- v2.0.5LatestSep 16, 2026
Score
40% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts
Declare a license in the repository or the registry entry.
Other Rust integrations
rust.wiki
wiki.rustOfficial publisher
Remote rust.wiki endpoint giving an assistant live Rust game data: item and skin prices, loot tables, monuments, servers and market stats.

Plane (Rust)
Published by GitHub account ninerf
Unofficial single-binary MCP server that lets an assistant read and write Plane project management data: work items, comments, cycles, modules, pages and more.

AGZ Rust MCP
Published by GitHub account agzyazilim
A local Rust development MCP server that lets an assistant inspect Cargo workspaces, read source, run checks and tests, and resolve crate docs.

rust-panosmcp
Published by GitHub account mechubsec
MCP server for Palo Alto Networks PAN-OS firewalls: read-only inspection plus scoped, audited, two-person-approved configuration changes.

rust-proxmoxmcp
Published by GitHub account mechubsec
Proxmox VE MCP server for many clusters, with guest protection and plan/approve/apply change control for destructive operations.