/ 目錄 / 演練場 / SkillForge
← 全部伺服器 ↗ GitHub
● 社群 tripleyak ⚡ 即開即用

SkillForge

作者 tripleyak · tripleyak/SkillForge

Phase-based skill router + creator — triages each request, analyzes via 11 thinking lenses, generates XML specs, and runs multi-agent approval before shipping.

SkillForge is a meta-skill for Claude Code and Codex that decides whether to USE an existing skill, IMPROVE one, CREATE new, or COMPOSE from multiple. It runs 4 phases: triage → deep 11-lens analysis → XML spec → generation with multi-agent synthesis. v5.0 streamlined the SKILL.md to 313 lines with lazy-loaded references so the context stays lean.

▶ 觀看演示 📦 安裝 💡 使用場景

為什麼要用

核心特性

即時演示

實際使用效果

skillforge-skill.replay ▶ 就緒
0/0

安裝

選擇你的客戶端

~/Library/Application Support/Claude/claude_desktop_config.json  · Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "skillforge-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/tripleyak/SkillForge",
        "~/.claude/skills/SkillForge"
      ],
      "_inferred": true
    }
  }
}

開啟 Claude Desktop → Settings → Developer → Edit Config。儲存後重啟應用。

~/.cursor/mcp.json · .cursor/mcp.json
{
  "mcpServers": {
    "skillforge-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/tripleyak/SkillForge",
        "~/.claude/skills/SkillForge"
      ],
      "_inferred": true
    }
  }
}

Cursor 使用與 Claude Desktop 相同的 mcpServers 格式。專案級設定優先於全域。

VS Code → Cline → MCP Servers → Edit
{
  "mcpServers": {
    "skillforge-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/tripleyak/SkillForge",
        "~/.claude/skills/SkillForge"
      ],
      "_inferred": true
    }
  }
}

點擊 Cline 側欄中的 MCP Servers 圖示，然後選 "Edit Configuration"。

~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "skillforge-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/tripleyak/SkillForge",
        "~/.claude/skills/SkillForge"
      ],
      "_inferred": true
    }
  }
}

格式與 Claude Desktop 相同。重啟 Windsurf 生效。

~/.continue/config.json
{
  "mcpServers": [
    {
      "name": "skillforge-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/tripleyak/SkillForge",
        "~/.claude/skills/SkillForge"
      ]
    }
  ]
}

Continue 使用伺服器物件陣列，而非映射。

~/.config/zed/settings.json
{
  "context_servers": {
    "skillforge-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/tripleyak/SkillForge",
          "~/.claude/skills/SkillForge"
        ]
      }
    }
  }
}

加入 context_servers。Zed 儲存後熱重載。

claude mcp add skillforge-skill -- git clone https://github.com/tripleyak/SkillForge ~/.claude/skills/SkillForge

一行命令搞定。用 claude mcp list 驗證，claude mcp remove 移除。

使用場景

實戰用法： SkillForge

Find the right existing skill instead of reinventing one

👤 Anyone with a big skill library and no memory of what's installed ⏱ ~5 min beginner

何時使用： You have a task and aren't sure if something already covers it.

步驟
  1. Describe the task
    Use skillforge. I need to turn our docs site into a searchable knowledge base.✓ 已複製
    → Triage returns USE_EXISTING with a recommendation (e.g. skill-seekers)
  2. Confirm and hand off
    Good — run that skill.✓ 已複製
    → Routes to the recommended skill

結果： You use what's already there rather than writing a new skill.

注意事項
  • Triage is over-eager to create new — Explicitly ask 'prefer USE_EXISTING unless match score < 70%'
搭配使用： skill-seekers-skill

Create a new skill with engineered rigor

👤 Skill authors who want quality, not vibes ⏱ ~30 min intermediate

何時使用： Triage returned CREATE_NEW and you want the full 4-phase generation.

步驟
  1. Approve triage
    skillforge — CREATE_NEW for a skill that drafts ADRs from design discussions.✓ 已複製
    → Phase 1 analysis begins with 11 lenses
  2. Review the spec
    Show me the XML spec before generation.✓ 已複製
    → Structured spec you can critique
  3. Run synthesis
    Generate, then run the multi-agent panel.✓ 已複製
    → Panel verdict with unanimous approval or concrete objections

結果： A production-quality SKILL.md ready to install.

注意事項
  • Panel stuck in endless loop — Cap iterations — SkillForge supports an iteration limit flag
搭配使用： skill-seekers-skill

組合

與其他 MCP 搭配，撬動十倍槓桿

skillforge-skill + skill-seekers-skill

Use SkillForge to decide, then Skill Seekers to actually ingest sources for CREATE_NEW

skillforge triage, then if CREATE_NEW, use skill-seekers to pull in the docs.✓ 已複製
skillforge-skill + filesystem

Skill output goes straight into ~/.claude/skills/

Write the generated SKILL.md and references to ~/.claude/skills/<name>/.✓ 已複製

工具

此 MCP 暴露的能力

工具輸入參數何時呼叫成本
Phase 0 triage task description + skill library Start of every request 0
Phase 1 deep analysis (11 lenses) task After triage, before generation 0
Phase 2-3 XML spec + generation analysis CREATE_NEW or IMPROVE_EXISTING 0
Phase 4 multi-agent synthesis draft Final quality gate 0

成本與限制

運行它的成本

API 配額
None
每次呼叫 Token 數
Full 4-phase generation is token-heavy (10-30k)
費用
Free — skill is local
提示
Use triage-only for simple routing; invoke full pipeline only for real skill authoring.

安全

權限、密鑰、影響範圍

憑證儲存： No credentials — skill is prompts
資料出站： None from the skill itself

故障排查

常見錯誤與修復

Triage picks wrong verdict

Provide the current skill library list so triage can match properly.

Multi-agent panel keeps rejecting

Check the objections — often a missing section in the spec. Fix once, re-run.

Generated skill feels generic

Give more source material in Phase 1; skill quality follows source quality.

替代方案

SkillForge 對比其他方案

替代方案何時用它替代權衡
skill-seekers-skillYou want document-ingest-first rather than analysis-firstSeekers is heavier on source processing; SkillForge is heavier on engineering rigor
Manual SKILL.md authoringThe skill is small and you have a clear specFastest for trivial skills

更多

資源

📖 閱讀 GitHub 上的官方 README

🐙 查看未解決的 issue

🔍 瀏覽全部 400+ MCP 伺服器和 Skills