/ 目錄 / 演練場 / skill-optimizer
● 社群 hqhq1025 ⚡ 即開即用

skill-optimizer

作者 hqhq1025 · hqhq1025/skill-optimizer

Audit a single SKILL.md using real session data plus static analysis — tells you exactly what to change.

skill-optimizer looks at a SKILL.md through two lenses: (1) session logs — did it activate when expected, was the output useful — and (2) static analysis — is the frontmatter clean, is the description properly intent-scoped, are examples calibrated. It outputs concrete edits you can apply. Pairs well with SkillCompass (which does bundle-level ranking).

為什麼要用

核心特性

即時演示

實際使用效果

skill-optimizer-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": {
    "skill-optimizer-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/hqhq1025/skill-optimizer",
        "~/.claude/skills/skill-optimizer"
      ],
      "_inferred": true
    }
  }
}

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

~/.cursor/mcp.json · .cursor/mcp.json
{
  "mcpServers": {
    "skill-optimizer-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/hqhq1025/skill-optimizer",
        "~/.claude/skills/skill-optimizer"
      ],
      "_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": {
    "skill-optimizer-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/hqhq1025/skill-optimizer",
        "~/.claude/skills/skill-optimizer"
      ],
      "_inferred": true
    }
  }
}

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

~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "skill-optimizer-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/hqhq1025/skill-optimizer",
        "~/.claude/skills/skill-optimizer"
      ],
      "_inferred": true
    }
  }
}

格式與 Claude Desktop 相同。重啟 Windsurf 生效。

~/.continue/config.json
{
  "mcpServers": [
    {
      "name": "skill-optimizer-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/hqhq1025/skill-optimizer",
        "~/.claude/skills/skill-optimizer"
      ]
    }
  ]
}

Continue 使用伺服器物件陣列,而非映射。

~/.config/zed/settings.json
{
  "context_servers": {
    "skill-optimizer-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/hqhq1025/skill-optimizer",
          "~/.claude/skills/skill-optimizer"
        ]
      }
    }
  }
}

加入 context_servers。Zed 儲存後熱重載。

claude mcp add skill-optimizer-skill -- git clone https://github.com/hqhq1025/skill-optimizer ~/.claude/skills/skill-optimizer

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: skill-optimizer

How to diagnose and fix one underperforming skill

👤 Skill authors iterating on a specific skill ⏱ ~30 min intermediate

何時使用: You notice one skill never activates, or activates wrongly.

前置條件
  • Session logs or transcripts — Claude Code stores traces locally; skill-optimizer reads them
  • Skill cloned — git clone https://github.com/hqhq1025/skill-optimizer ~/.claude/skills/skill-optimizer
步驟
  1. Point it at the skill
    Analyze ~/.claude/skills/my-skill/SKILL.md using my last 2 weeks of session logs.✓ 已複製
    → Static + runtime findings
  2. Review the patch
    Show me the proposed SKILL.md diff.✓ 已複製
    → Minimal, reviewable diff
  3. Apply and validate
    Apply the patch and tell me what to watch for this week.✓ 已複製
    → Validation checklist

結果: A targeted improvement with an accountability plan.

注意事項
  • Optimizing for activation rate but breaking scope — Include a 'false positive rate' acceptance metric
搭配使用: skillcompass-skill

Polish your first SKILL.md before publishing

👤 First-time skill authors ⏱ ~20 min beginner

何時使用: You wrote a SKILL.md and want expert review before shipping.

步驟
  1. Ask for a review
    Review my SKILL.md against skill-optimizer's static checks.✓ 已複製
    → Checklist with pass/fail + rationale
  2. Fix issues
    Apply the fixes and show me the final version.✓ 已複製
    → Clean SKILL.md

結果: A ship-ready SKILL.md.

注意事項
  • Over-promising in the description — Be specific about what the skill does; Claude uses it to decide activation

組合

與其他 MCP 搭配,撬動十倍槓桿

skill-optimizer-skill + skillcompass-skill

SkillCompass ranks the bundle; skill-optimizer dives into the worst offender

Use SkillCompass to pick the weakest skill, then optimize its SKILL.md.✓ 已複製
skill-optimizer-skill + filesystem

Apply SKILL.md patches directly

Apply the proposed diff to ~/.claude/skills/my-skill/SKILL.md.✓ 已複製

工具

此 MCP 暴露的能力

工具輸入參數何時呼叫成本
static-analysis SKILL.md path Before publishing or during iteration 0
log-based-analysis logs + skill path Post-use tuning 0
patch-generation findings After analysis 0

成本與限制

運行它的成本

API 配額
none beyond LLM
每次呼叫 Token 數
2–10k per skill analysis
費用
free
提示
Static analysis is nearly free; log analysis costs more — run it only when iterating

安全

權限、密鑰、影響範圍

憑證儲存: none; logs may contain sensitive prompts — treat their path carefully
資料出站: none

故障排查

常見錯誤與修復

No logs found

Confirm Claude Code trace path; pass the path explicitly if non-default.

Suggestions conflict with intended scope

Tell the optimizer your scope constraints; it can respect non-goals

替代方案

skill-optimizer 對比其他方案

替代方案何時用它替代權衡
skillcompass-skillYou need bundle-level ranking, not single-skill depthBreadth over depth

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