SkillDB

io.github.devchad-cmd由 GitHub 帳號 devchad-cmd 發布v0.8.4-registry.1更新於 Nov 3, 2026

Search and load AI agent skills. Free metadata; full content requires a Pro or Studio account.

已驗證Streamable HTTP可網頁執行知識與記憶開發者工具

概覽

AI 產生的概覽

SkillDB 讓 AI 助理搜尋、預覽並載入 5900 多個專家技能,付費方案下還可管理私有技能。

可以做什麼
  • 依關鍵字、分類或技能包搜尋、預覽與列出技能
  • 載入完整技能內容,並為專案取得推薦與建議
  • 管理私有技能:列出自己擁有與共享的技能,建立與更新(需 Studio 方案與寫入權限金鑰)
  • 在工作階段中設定或更換 API 金鑰,並清除本機快取的技能
適用情境
適合希望編碼助理取得針對特定任務的精選指令(例如程式碼審查、React 模式、安全強化)而非只依賴一般知識的情境,也適合維護自有私有技能庫的使用者。
執行需求
可使用 skilldb.dev 的託管端點(Streamable HTTP,OAuth 或 API 金鑰),或以 npx 執行本機 npm 套件,需 Node.js 18 以上。僅瀏覽中繼資料時 API 金鑰為選用;完整技能內容需 Pro 或 Studio 金鑰,寫入私有技能需具備寫入權限的 Studio 金鑰。金鑰可透過 SKILLDB_API_KEY 環境變數或 .skilldbrc 檔案提供。
安裝前請注意
API 金鑰屬於憑證,可能經由命令列參數傳入或存放在 .skilldbrc 檔案中,請勿放入共享儲存庫。完整內容與私有技能需要付費的 Pro 或 Studio 帳戶。具寫入權限的金鑰可建立與更新私有技能,purge 與 remove 指令會刪除本機快取的技能。匿名目錄端點只回傳中繼資料與預覽。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 SkillDB,將其新增到工作區。
  2. 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
  3. 為需要使用其工具的對話啟用該服務。

Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/io.github.devchad-cmd%2Fskilldb/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。

{
  "mcpServers": {
    "skilldb": {
      "type": "http",
      "url": "https://skilldb.dev/api/mcp"
    }
  }
}

README

skilldb

CLI, TypeScript SDK, and MCP Server for SkillDB — discover, install, and manage AI agent skills.

5,900+ expert skills across 429 packs for Claude Code, Cursor, Windsurf, VS Code, and any MCP-compatible AI tool.

[Add SkillDB to Cursor] [Install in VS Code] [npm]

Open source for transparency. This is the full source for the skilldb npm package — the CLI and MCP server that run on your machine and handle your API key. It's published to npm via Trusted Publishing (OIDC, with provenance), so the published artifact is verifiably built from this repo. Audit away, and PRs welcome. MIT licensed.

MCP Server

Connect SkillDB directly to your AI coding tool via the Model Context Protocol:

bash
# Step 1 (optional): install globally. You can also skip this — the npx form below works without it.npm install -g skilldb
# Step 2 (optional): Get a Pro or Studio API key for full content at https://skilldb.dev/api-access
# Step 3: Add to Claude Code (omit --api-key for metadata-only access)claude mcp add skilldb -- npx -p skilldb skilldb-mcp --api-key sk_live_YOUR_KEY

Note: skilldb-mcp is a binary inside the skilldb package, not its own npm package — always invoke it as npx -p skilldb skilldb-mcp (not npx skilldb-mcp).

⚠️ Without an API key, you can search and browse skills (metadata only). With a Pro or Studio API key, you get full skill content — the actual markdown instructions your agent uses. Get your key in 30 seconds at skilldb.dev/api-access.

Cursor — add to .cursor/mcp.json:

json
{  "mcpServers": {    "skilldb": {      "command": "npx",      "args": ["-p", "skilldb", "skilldb-mcp"]    }  }}

Windsurf — add to mcp_config.json:

json
{  "mcpServers": {    "skilldb": {      "command": "npx",      "args": ["-p", "skilldb", "skilldb-mcp", "--api-key", "sk_live_xxx"]    }  }}

VS Code — one-liner:

bash
code --add-mcp '{"name":"skilldb","command":"npx","args":["-p","skilldb","skilldb-mcp"]}'

Or add the same server block to .vscode/mcp.json (workspace) / user mcp.json under "servers".

Once connected, your AI assistant can search and load skills natively. Just ask:

  • "Search SkillDB for React performance patterns"
  • "Load the code review skill"
  • "What skills should I use for this project?"
  • "Save this as a private skill" (Studio plan — see below)

10 tools exposed: skilldb_search, skilldb_get, skilldb_list, skilldb_suggest, skilldb_recommend, skilldb_purge, skilldb_set_key (set/swap your API key mid-session, no restart), and the private-skill tools skilldb_my_skills, skilldb_create_skill, skilldb_update_skill (Studio plan + write-scoped key).

Hosted MCP endpoints

  • Authenticated integration: https://skilldb.dev/api/mcp uses Streamable HTTP with OAuth or a SkillDB API key. Full skill content requires a Pro or Studio account.
  • Anonymous SkillDB Catalog: https://skilldb.dev/api/mcp/catalog uses Streamable HTTP without a SkillDB account. Its two tools, skilldb_search and skilldb_get_preview, return public metadata and short previews only. It does not retrieve full skill content, install or execute skills, link accounts, or access private/team libraries.

Quick Start (CLI)

bash
# Search for skills (with inline preview)npx skilldb search "code review"
# Install a skill packnpx skilldb add software-skills
# Activate a smart profile (only loads what you need)npx skilldb use frontend
# Or auto-detect from your projectnpx skilldb use auto

Install

bash
npm install -g skilldb

Or use npx skilldb without installing.

The Memory Problem (And How We Solve It)

Loading 500 skills into your agent's context is worse than loading the right 5. SkillDB uses a 3-layer architecture to manage this:

.skilldb/├── skills/          ← Full cache (everything downloaded)├── active/          ← Active profile (what your agent sees — 5-15 focused skills)├── slim/            ← Compressed cheat sheets (10x smaller)├── config.json      ← Profile + budget settings└── manifest.json    ← What's installed

Your CLAUDE.md or .cursorrules points at .skilldb/active/, not the full cache. The agent gets focused, relevant skills — not noise.

CLI Commands

Core

skilldb init

Detect your IDE (Claude Code, Cursor, Codex CLI, OpenClaw), create .skilldb/, and add integration config.

bash
skilldb init
skilldb search <query>

Search skills by keyword. Shows inline slim preview so you know what you're getting before downloading.

bash
skilldb search "debugging"skilldb search "api design" --category "Technology & Engineering"skilldb search "testing" --limit 10
skilldb add <pack>

Download a skill pack to local cache. Idempotent — skips already-cached skills.

bash
skilldb add software-skillsskilldb add autonomous-agent-skillsskilldb add vibe-coding-security-skills
skilldb get <id>

Download a single skill.

bash
skilldb get software-skills/code-review
skilldb info <id>

Show metadata, slim summary, and full preview for a skill.

bash
skilldb info software-skills/code-review
skilldb list

List available categories, packs, and skills.

bash
skilldb listskilldb list --category "Autonomous Agents"skilldb list --pack software-skills
skilldb login

Prompts for your API key and saves it to ~/.skilldbrc (user-wide) so every project picks it up via the resolution chain in Authentication. The recommended alternative to setting SKILLDB_API_KEY by hand.

bash
skilldb login

Smart Loading

skilldb use <profile>

Activate a focused profile — only loads relevant skills into .skilldb/active/.

bash
skilldb use frontend          # React, testing, web-polish, accessibilityskilldb use backend           # API design, databases, security, performanceskilldb use devops            # Kubernetes, Docker, CI/CD, monitoringskilldb use security          # Trust audit, input validation, credentialsskilldb use data              # SQL, pipelines, analytics, visualizationskilldb use fullstack         # Frontend + backend combinedskilldb use mobile            # Mobile / React Native / app developmentskilldb use ai-agent          # Autonomous agent meta-skillsskilldb use auto              # Auto-detect from your project filesskilldb use --list            # Show available profilesskilldb use --current         # Show active profileskilldb use none              # Deactivate (clear active/)

auto mode scans your package.json, file extensions, and imports to recommend the right profile.

skilldb budget

Set a maximum context budget so your agent doesn't get overloaded.

bash
skilldb budget                # Show current usageskilldb budget set 5000       # Max 5,000 linesskilldb budget set 50k        # Max 50,000 tokensskilldb budget optimize       # Re-rank and trim active skills to fit
skilldb slim

Generate compressed cheat-sheet versions of skills (~30 lines instead of 300).

bash
skilldb slim                  # Slim all active skillsskilldb slim software-skills  # Slim a specific packskilldb slim --ratio 0.3      # Keep 30% of content

Management

skilldb update

Check for newer versions of installed skills and update them.

bash
skilldb update                # Update everythingskilldb update software-skills  # Update one pack
skilldb remove

Uninstall skills from local cache.

bash
skilldb remove software-skills/code-review    # Remove single skillskilldb remove software-skills                # Remove entire packskilldb remove --unused                       # Remove skills not in active profile
skilldb doctor

Health check and audit of your skill setup.

bash
skilldb doctor

Shows: outdated skills, unused skills, missing dependencies, budget status, coverage gaps.

skilldb stats

Local statistics dashboard.

bash
skilldb stats

Shows: installed count, total lines, tokens, active profile, categories covered.

skilldb diff <id>

Compare your local (possibly customized) version with the latest remote.

bash
skilldb diff software-skills/code-reviewskilldb diff                  # Diff all installed
skilldb purge

Remove cached skills to free disk space.

bash
skilldb purge --inactive      # Remove skills not in the active profileskilldb purge --all           # Remove ALL cached skills (including active)skilldb purge --slim          # Also drop slim summariesskilldb purge --dry-run       # Preview what would be removed, delete nothing

Private Skills (Studio)

Create skills that only you (and people you share them with) can read. They surface in your own search results when you're authenticated. Creating/editing requires the Studio plan and an API key with write scope — mint one at skilldb.dev/account#api-keys.

skilldb push <file.md>

Create (or update) a private skill from a local Markdown file. The skill name defaults to the filename; the title resolves as --title flag → first Markdown # H1 → filename.

bash
skilldb push ./my-react-patterns.md                    # create (pack defaults to "personal")skilldb push ./my-react-patterns.md -p team --tags react,perfskilldb push ./my-react-patterns.md --update           # update the skill with the same name/pack

Options: -p, --pack, -n, --name, -t, --title, -d, --description, --tags <comma,list>, -u, --update.

skilldb my-skills

List your private skills (ones you own plus skills shared with you).

bash
skilldb my-skills

You can also do all of this from your agent via the MCP tools skilldb_create_skill / skilldb_update_skill / skilldb_my_skills.

Sharing & Export

skilldb export

Export your setup for sharing or IDE integration.

bash
skilldb export claude         # CLAUDE.md snippet with skill referencesskilldb export cursor         # .cursorrules blockskilldb export profile        # Shareable .skilldb-profile.jsonskilldb export inject <id>    # Print skill to stdout (for piping)
skilldb recommend

Scan your project and get personalized skill recommendations.

bash
skilldb recommend             # Analyze project and suggestskilldb recommend --install   # Auto-install recommendations

SDK Usage

typescript
import { createClient } from 'skilldb';
const db = createClient(); // auto-loads key from env/config
// Search with slim previewsconst results = await db.search('code review');for (const skill of results.skills) {  console.log(skill.title);  console.log(skill.slim); // Quick summary}
// Get full skill content (requires a configured Pro or Studio API key)const skill = await db.get('software-skills/code-review.md');console.log(skill.content);
// List with filtersconst listing = await db.list({ category: 'Autonomous Agents', limit: 10 });

Authentication

The SDK and CLI resolve your API key in order:

  1. SKILLDB_API_KEY environment variable (CI-friendly)
  2. .skilldbrc in project root (project-specific)
  3. ~/.skilldbrc in home directory (user-wide)

Access tiers:

  • No key — browse and search (metadata only).
  • Pro or Studio key — full skill content (the actual markdown instructions).
  • Studio key with write scope — private skills (skilldb push / my-skills, and the create/update MCP tools). Read-only keys are rejected with a write-scope error. Create a write-scoped key at skilldb.dev/account#api-keys.

Profiles

Built-in profiles map to curated skill sets:

ProfileSkills loadedBest for
frontendreact-patterns, web-polish, testing, accessibilityReact/Vue/Angular apps
backendapi-design, databases, security, performanceNode/Python/Go services
devopskubernetes, docker, ci-cd, monitoring, cloudInfrastructure & deployment
securitytrust-audit, input-validation, credentials, hardeningSecurity review & hardening
datasql, pipelines, analytics, visualizationData engineering & analysis
fullstackfrontend + backend combinedFull-stack applications
mobilemobile patterns, app architecture, UIMobile / React Native apps
ai-agentautonomous-agent-skills, task-decomposition, planningBuilding AI agents
autoDetected from projectAny project

Requirements

  • Node.js >= 18

License

MIT

來源:README.md,提交 a4459e4

工具

0
該伺服器需要登入,因此 SourceWeft 無法在沒有帳號的情況下讀取其工具,也未收錄。已連接帳號所能看到的內容可能因帳號及其權限而異。

版本歷史

2
  1. v0.8.4-registry.1最新Oct 6, 2026
  2. v0.8.3Sep 16, 2026

評分

已滿足適用檢查項的 85%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分