🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman
Skill server and `skb` CLI for agent skills
Skill Base gives teams one place to publish and manage Skills, then install or update them into the folders each agent expects. It tracks versions, permissions, collections, and local installs so people do not have to copy Markdown by hand.

Builders who want one shared place to publish and sync Skills for Cursor, Claude Code, Codex, and similar agents.
You can keep team Skills current across agents without copying files or guessing which version is latest.
What it does
Publish skills
Upload Skills through the web app, `skb publish`, or GitHub import from public repos.
Install and update skills
Use `skb install`, `skb update`, and `skb delete` to manage local Skill copies in the right agent folders.
Manage collections
Install curated packs with `skb install --collection <id-or-slug>` and browse collections in the UI.
Control visibility and access
Mark Skills as public or private and manage owner, collaborator, and user permissions.
Browse versions and changelogs
Search Skills, switch versions, and review changelogs, tags, and favorites in the web UI.
Run a desktop client
Use the native desktop app for a local market and Skill management.
How to get it
- 1Server (Node.js >= 18)
npx skill-base -d ./skill-data -p 8000
- 2CLI
npm install -g skill-base-cli skb init -s http://localhost:8000 skb search vue skb install some-skill --ide cursor skb install --collection 1 --ide cursor skb install --collection frontend-team --ide cursor skb login # required for publish skb publish ./my-skill --changelog "First release"
README
Skill Base
English | 中文
Private distribution for Agent Skills. Publish once; install, update, and roll back across Cursor, Claude Code, Codex, OpenClaw, and similar assistants — one small server, one CLI (
skb).
Why
Many teams already have dozens or hundreds of Skills, but they are not team assets yet — just Markdown scattered across chat logs, personal machines, and project folders. People download from group chats, copy a colleague's old copy, or edit their local file directly. Over time, no one can say which copy is current, who changed what, or which projects still run an old version.
Some teams move Skills into Git, which helps: you get commits and a single repo. But Git answers "where do files live?", not "how does the team use them?" It does not fix fragmented Skill paths across Cursor, Claude Code, Qoder, and other agents. The real pain is distribution, install, updates, and access:
- Distribution by hand — zip files, pasted Markdown, new hires asking "where is the latest?"
- Versions by guesswork —
final,final2,latesteverywhere; local edits untracked - Paths by memory — Cursor, Claude Code, Qoder, etc. use different Skill folders; install and update depend on tribal knowledge
- Access blocks PM/QA — they need the same standards but should not need repo access or Git workflows for one Skill
- Updates stop at one machine or project — one person fixes a Skill; other projects and teammates keep the old copy
Skill Base turns Skills into publishable, versioned, and installable team assets.
What it does
| Capability | How |
|---|---|
| Publish | Web upload, skb publish, or GitHub import from public repos |
| Install / update / delete | skb install / skb update / skb delete with IDE Skill paths and local install tracking; skb install --collection <id-or-slug> for curated packs |
| Browse | Web UI for search, version switching, changelogs, tags, collections (max 10 skills), favorites |
| Desktop | Native desktop client — download |
| Visibility | public / private skills; owner / collaborator / user permissions |
Quick start
Server (Node.js >= 18):
npx skill-base -d ./skill-data -p 8000
Open the URL, complete first-time setup, create an admin account.
CLI:
npm install -g skill-base-cli
skb init -s http://localhost:8000
skb search vue
skb install some-skill --ide cursor
skb install --collection 1 --ide cursor
skb install --collection frontend-team --ide cursor
skb login # required for publish
skb publish ./my-skill --changelog "First release"
Details: Getting Started · CLI: docs/zh/cli.md (中文)
Design philosophy
- Skills are the unit — not a generic file dump or agent orchestrator.
- SQLite + ZIP on disk —
skills.dbindexes; each release is a versioned archive. Backup = copy one directory. - One Node process — no Redis, MySQL, queue, or microservice mesh for a problem this size.
Longer write-up: Architecture.
Documentation
| Doc | Contents |
|---|---|
| Getting Started | Server, CLI, first install/publish |
| Usage | Web UI, desktop app, GitHub import, OpenClaw prompts, Skill format |
| Deployment | Docker, PM2, flags, sessions, backup |
| Architecture | Data model and stack choices |
| Desktop | Download and features |
| CLI | skb command reference (中文) |
| API | HTTP API (中文) |
| Changelog | Server / npm release notes |
| Author | Maintenance contract, Author's Lab |
| Docs index | Bilingual documentation map |
中文入口:docs/zh/README.md
Author's Lab
Related projects from the author.
| skillbase.reaidea.com | Hosted Skill Base |
| reaidea.com | Author homepage and future tools |
More entries land in docs/author.md when they ship.
Maintenance
This repo is maintained by a part-time independent developer on evenings and weekends.
Issues and PRs are welcome. There is no on-call SLA; clear, scoped PRs are easier to merge than open-ended feature asks.
Full note: docs/author.md.
License
Files in the repo
- .github
- bin
- cli
- data
- data2
- demo
- desktop-tauri
- docs
- e2e-tests
- skills
- src
- static
- test
- tests
- vendor
- web
- website
- .cursorignore
- .dockerignore
- .gitignore
- AGENTS.md
- CHANGELOG.md
- design.md
- Dockerfile
- package-lock.json
- package.json
- pnpm-lock.yaml
- README.md
- website.zip
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More tools
The best-benchmarked open-source AI memory system. And it's free.
Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and remote runtime.

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io
Never stop coding. Free MIT AI gateway: one endpoint, 352 providers (150+ free), 1200+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 550+ contributors
Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.