/ 目錄 / 演練場 / gsd-skill-creator
● 社群 Tibsfox ⚡ 即開即用

gsd-skill-creator

作者 Tibsfox · Tibsfox/gsd-skill-creator

Observes your Claude sessions for 3+ repeated tool patterns and suggests new skills — combats context rot, session amnesia, and missing workflow memory.

Guided Software Development (GSD) is an adaptive learning framework for Claude Code with an observe -> detect -> suggest -> apply -> learn -> compose loop. It watches tool sequences, detects patterns appearing 3+ times, proposes skill creation (with your explicit approval), auto-loads relevant skills within token budgets, learns from corrections (3-correction minimum, 7-day cooldown), and composes agent teams when skills co-activate. Ships with a retro-UI desktop app built on Tauri.

為什麼要用

核心特性

即時演示

實際使用效果

gsd-skill-creator-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": {
    "gsd-skill-creator-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/Tibsfox/gsd-skill-creator",
        "~/.claude/skills/gsd-skill-creator"
      ],
      "_inferred": true
    }
  }
}

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

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

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

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

格式與 Claude Desktop 相同。重啟 Windsurf 生效。

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

Continue 使用伺服器物件陣列,而非映射。

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

加入 context_servers。Zed 儲存後熱重載。

claude mcp add gsd-skill-creator-skill -- git clone https://github.com/Tibsfox/gsd-skill-creator ~/.claude/skills/gsd-skill-creator

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: gsd-skill-creator

Let GSD detect your repeating workflow and codify it as a skill

👤 Heavy Claude Code users who keep redoing the same steps ⏱ ~30 min intermediate

何時使用: You've done 'clone repo -> run tests -> fix lint -> open PR' enough times that muscle memory should be code.

前置條件
  • Skill cloned — git clone https://github.com/Tibsfox/gsd-skill-creator ~/.claude/skills/gsd-skill-creator
步驟
  1. Run with observation mode on
    Enable gsd-skill-creator in observe mode. Run my normal PR flow a few times over the next week.✓ 已複製
    → Silent logging of tool sequences
  2. Review suggestions
    Show me detected patterns with 3+ occurrences, ranked by confidence.✓ 已複製
    → Ranked list with proposed skill names
  3. Confirm and let it build
    Promote pattern #1 into a skill — approve generation.✓ 已複製
    → A new skill appears in ~/.claude/skills/

結果: Your actual workflow becomes a reusable skill, grounded in data.

注意事項
  • Approving every suggestion — Promote only the top 1–2 — skill bloat hurts the picker
搭配使用: skill-forge-skill

Compose co-activating skills into an agent team

👤 Engineers whose skills consistently fire together ⏱ ~30 min advanced

何時使用: After a month of use, the skill picker keeps selecting the same 3 skills in sequence.

步驟
  1. Review co-activation report
    Show skill co-activation report — which skills fire within the same task 80%+ of the time?✓ 已複製
    → Cluster diagram
  2. Compose team
    Compose the top cluster into a named agent team.✓ 已複製
    → Team definition that orchestrates the cluster

結果: A higher-level orchestrator that triggers the full cluster in one shot.

Resume a long project without repeating old mistakes

👤 Teams doing multi-week Claude-driven work ⏱ ~15 min beginner

何時使用: You're picking up a project Claude has lost context on.

步驟
  1. Resume with skill history
    Resume project X using gsd-skill-creator memory — what patterns and corrections have I applied so far?✓ 已複製
    → Summary of prior skills and corrections

結果: Claude starts where it left off, not from zero.

組合

與其他 MCP 搭配,撬動十倍槓桿

gsd-skill-creator-skill + skill-forge-skill

GSD detects the pattern; skill-forge scaffolds the formal skill

Hand the top detected pattern to skill-forge build for proper structure.✓ 已複製
gsd-skill-creator-skill + planning-with-files-skill

GSD observes sessions; planning-with-files persists state across them

Use both — GSD for pattern learning, planning-with-files for session state.✓ 已複製

工具

此 MCP 暴露的能力

工具輸入參數何時呼叫成本
observe_start session id or none Start of the adoption period 0
detect_patterns min_occurrences? Weekly review 0
promote_to_skill pattern id, name After user approval 0
compose_team cluster id When clusters are stable 0
learn_from_correction correction context After 3+ corrections in cooldown window 0

成本與限制

運行它的成本

API 配額
None
每次呼叫 Token 數
Observation is cheap; generation phases are moderate
費用
Free, BSL 1.1 (converts to GPL 3.0 in 2030)
提示
Only run detection weekly, not per-session

安全

權限、密鑰、影響範圍

憑證儲存: Observation data is local
資料出站: None by default

故障排查

常見錯誤與修復

No patterns detected after weeks of use

Your sessions may vary too much; loosen the pattern-matching threshold

GSD-OS desktop app won't launch

Rebuild Tauri bundle for your platform; check Rust toolchain

驗證: cargo --version

替代方案

gsd-skill-creator 對比其他方案

替代方案何時用它替代權衡
skill-forgeYou already know what skill you want — skip the detection layerNo auto-detection
Manual SKILL.md authoringOne-off skillNo adaptive learning

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