/ 目錄 / 演練場 / swing-skills
● 社群 TheStack-ai ⚡ 即開即用

swing-skills

作者 TheStack-ai · TheStack-ai/swing-skills

6 cognitive firewalls for AI agents — clarify, research, options, review, trace, mortem — each targeting a named bias: premature closure, hallucination, anchoring, confirmation, black-box reasoning, optimism.

A skill pack of 6 self-contained cognitive tools that plug specific biases. swing-clarify forces 5W1H on ambiguous requests. swing-research runs a 4-stage verified pipeline with S/A/B/C source grading. swing-options generates 5 probability-weighted alternatives including unconventional ones. swing-review does structured devil's advocacy (steel-man first, then 3-vector attack). swing-trace inventories assumptions and surfaces weakest links. swing-mortem does prospective failure analysis.

為什麼要用

核心特性

即時演示

實際使用效果

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

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

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

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

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

格式與 Claude Desktop 相同。重啟 Windsurf 生效。

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

Continue 使用伺服器物件陣列,而非映射。

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

加入 context_servers。Zed 儲存後熱重載。

claude mcp add swing-skill -- git clone https://github.com/TheStack-ai/swing-skills ~/.claude/skills/swing-skills

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: swing-skills

Force clarification before an ambiguous task is executed

👤 Anyone who's seen AI agents gallop off in the wrong direction ⏱ ~10 min beginner

何時使用: A stakeholder dropped a short, ambiguous request.

前置條件
  • Skill installed — git clone https://github.com/TheStack-ai/swing-skills ~/.claude/skills/swing-skills
步驟
  1. Run swing-clarify
    Use swing-clarify. Request: 'make the dashboard better'. Decompose 5W1H.✓ 已複製
    → What/who/when/where/why/how decomposition + clarifying questions

結果: A concrete scope you can actually execute against.

搭配使用: agent-skill-tdd-skill

Research a topic with graded sources, not a wall of links

👤 Analysts, researchers ⏱ ~45 min intermediate

何時使用: You want a defensible brief, not hallucinated bullet points.

步驟
  1. Run swing-research
    Use swing-research. Question: 'what's the state of Rust in web backends (2025)?' Output claims with S/A/B/C grades.✓ 已複製
    → Claim table with tier + source

結果: A brief you could defend to a skeptical stakeholder.

注意事項
  • B/C-tier claims treated as fact — The grade is there to weight; read them as weaker
搭配使用: firecrawl · brave-search

Break anchoring by generating five weighted alternatives

👤 Decision-makers who notice they're anchoring ⏱ ~20 min beginner

何時使用: You've got a favorite solution and want a sanity check.

步驟
  1. Run swing-options
    Use swing-options. Decision: which DB for our new service? Give 5 probability-weighted options including unconventional.✓ 已複製
    → 5 options with weights + rationale

結果: A decision made from options, not one default.

Structured review: steel-man then attack from 3 vectors

👤 Teams wanting a real review, not rubber-stamps ⏱ ~30 min intermediate

何時使用: Before committing to a plan or design.

步驟
  1. Run swing-review
    Use swing-review on design.md. Steel-man it first, then attack from correctness, operability, and cost.✓ 已複製
    → Strong version of the design, then 3 attack surfaces

結果: A design that either holds or gets better.

Run a pre-mortem on a plan before execution

👤 PMs and leads about to green-light something important ⏱ ~30 min intermediate

何時使用: Right before kickoff on risky work.

步驟
  1. Run swing-mortem
    Use swing-mortem. Assume plan.md failed in 6 months. List 5 failure scenarios with leading indicators.✓ 已複製
    → 5 scenarios + early-warning signals

結果: Early-warning instrumentation built in from day one.

Expose the weakest assumption in a decision tree

👤 Engineers debugging reasoning, not just code ⏱ ~20 min intermediate

何時使用: A decision feels off but you can't articulate why.

步驟
  1. Run swing-trace
    Use swing-trace on this decision: 'we'll use Kafka'. Inventory assumptions; call out the weakest.✓ 已複製
    → Assumption list ranked by risk

結果: The weakest link exposed before it breaks.

組合

與其他 MCP 搭配,撬動十倍槓桿

swing-skill + skills-skill-6

Layer Swing's firewalls with InfraNodus graph analysis

After swing-review, run InfraNodus critical-perspective for graph-based critique.✓ 已複製
swing-skill + agent-skill-tdd-skill

Clarify + research before the TDD flow

Start with swing-clarify + swing-research, then hand off to agent-skill-tdd.✓ 已複製

工具

此 MCP 暴露的能力

工具輸入參數何時呼叫成本
swing-clarify ambiguous request Any vague task 0
swing-research question Research with defensibility 0
swing-options decision Anchoring risk 0
swing-review plan/design Pre-commit review 0
swing-trace decision Reasoning audit 0
swing-mortem plan Before kickoff 0

成本與限制

運行它的成本

API 配額
None
每次呼叫 Token 數
Moderate per skill (research is highest)
費用
Free
提示
Use swing-clarify by default; others on-demand

安全

權限、密鑰、影響範圍

憑證儲存: None
資料出站: None beyond your LLM calls

故障排查

常見錯誤與修復

swing-research cites sources that don't check out

Cross-check S-tier sources manually; lower grades are weaker by design

swing-review feels like a formality

Force adversarial framing: 'what would an angry senior engineer say?'

替代方案

swing-skills 對比其他方案

替代方案何時用它替代權衡
InfraNodus skillsYou want graph-based thinking augmentationNo bias-specific firewalls
Manual reviewOne-off trivial decisionsNo structure

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