SuperAPI AI Directory

troubleshoot

Deployment & operations

View repository

Updated 10-15 18:46

About this skill

Troubleshoot Skill

A skill that guides diagnosis and resolution when problems occur. It helps VibeCoder solve problems without technical knowledge.


Trigger phrase

This skill is automatically triggered by the following phrases:

  • "It doesn't work", "I got an error", "It broke"
  • "It's not going well", "It failed"
  • "Why?", "What's the cause?"
  • "it's broken", "doesn't work", "error", "why?"

Overview

When a problem occurs, VibeCoder does not need to understand the technical details. This skill automatically diagnoses and presents solutions.


Diagnosis Flow

Step 1: Confirming the Symptoms

🔍 What happened?

Please tell me briefly:

  • "The screen is all white"
  • "The button doesn't work"
  • "The data isn't saved"

Step 2: Automatic Diagnosis

# 環境チェック
node -v
npm -v
git status -sb

# エラーログ確認
npm run build 2>&1 | tail -20
npm test 2>&1 | tail -20

# 依存関係チェック
npm ls --depth=0

Step 3: Identifying the Problem Category

Category Symptom Automatic Handling
Dependencies Cannot find module npm install
Type errors Type error error-recovery agent
Build error Build failed error-recovery agent
Runtime Screen doesn't display Log analysis
Environment settings Connection errors Environment variable check
CI/CD CI failed, pipeline failure dispatch to ci skill

CI/CD problems: If CI failures, test failures, or pipeline problems are detected,ci skill (skills/ci/), the process is delegated. The ci skill uses the ci-cd-fixer agent to perform automatic diagnosis and fixes.


Handling by Problem

"The button doesn't work / the form isn't submitted / the UI is broken"

For UI defects,reproduce and observe in the screen → fix → re-verifyis the fastest approach.

  1. Use agent-browser with the highest priority(Install: npm install -g agent-browser)
    # ページを開いてスナップショット取得
    agent-browser open https://example.com/target-page
    agent-browser snapshot -i -c
    
    # 要素参照でクリック・入力
    agent-browser click @e1
    agent-browser fill @e2 "test"
    
    • Reproduce at the target URL → narrow down the cause based on snapshots/console
    • Check the source code (UI/state management/API/validation) and fix
    • Confirm that it does not reproduce with the same steps
    • Reference: docs/OPTIONAL_PLUGINS.md
  2. Fallback when agent-browser is unavailable
    • MCP browser tools (chrome-devtools, playwright)
    • Reproduction steps (URL/steps/expected/actual)
    • Screenshot/video
    • Console logs/network logs

"The screen is all white"

🔍 診断中...

考えられる原因:
1. ビルドエラー
2. JavaScript エラー
3. ルーティング問題

🔧 自動チェック:
- ビルドログを確認... ✅ エラーなし
- コンソールエラーを確認... ❌ エラー発見

💡 解決策:
「直して」と言ってください。自動修正を試みます。

"npm install fails"

🔍 診断中...

エラー: `ERESOLVE unable to resolve dependency tree`

🔧 解決策:
1. キャッシュをクリア
2. node_modules を再インストール

実行してもいいですか?(はい/いいえ)

"Git is acting weird"

🔍 診断中...

検出: マージコンフリクト発生

🔧 解決策:
1. コンフリクト箇所を表示
2. 解決方法を提案

「解決して」と言えば自動マージを試みます。

Response Patterns for VibeCoder

Pattern 1: Automatically fixable

⚠️ 問題を検出しました

**問題**: モジュールが見つかりません
**原因**: 依存関係の未インストール

🔧 **自動修正します**
→ `npm install` を実行中...
✅ 修正完了!もう一度試してください。

Pattern 2: Confirmation needed

⚠️ 問題を検出しました

**問題**: 設定ファイルの変更が必要
**影響**: 動作に影響する可能性

🤔 **確認させてください**:
設定を変更しても大丈夫ですか?

→ 「変更して」で実行
→ 「説明して」で詳細を確認

Pattern 3: Escalation

⚠️ 複雑な問題を検出しました

**問題**: {{問題の概要}}
**試した修正**: 3回失敗

🆘 **Cursor (PM) への相談をおすすめします**

以下を Cursor に共有してください:
- エラー内容: {{要約}}
- 試した対応: {{リスト}}
- 推定原因: {{分析}}

「報告書を作って」で共有用レポートを生成します。

Automatic Responses to Common Questions

Question Automatic Response
"Why did it error?" Analyze the error log and explain the cause
"How can I fix it?" Present concrete resolution steps
"It was working before" Check recent changes with git log
"The same error over and over" Analyze the pattern and propose a fundamental fix

Diagnostic Commands

Quick Diagnostics

「診断して」
→ 全般的な健全性チェック
→ 問題があれば報告
→ なければ「問題なし」

Detailed Diagnostics

「詳しく診断して」
→ 依存関係チェック
→ ビルドテスト
→ テスト実行
→ 環境変数確認
→ 詳細レポート出力

Claude Code-specific diagnostics (CC 2.1.49+)

For complex session issues, the /debug command is effective.

Symptom Recommended
"The session is acting weird" /debug for session diagnostics
"MCP doesn't work" /debug to check MCP status
"It fails no matter how many times I try" /debug to output detailed logs
"The context is strange" /debug to retrieve context information
"Authentication error" claude auth status to check authentication status (CC 2.1.41+)
"The agent doesn't work" claude agents list to check agent definitions (CC 2.1.50+)
"The skill operates with old content" /clear to run and reset the skill cache (CC v2.1.63+)

Agent diagnostics (CC 2.1.50+)

CC 2.1.50 で claude agents CLI commands have been added:

claude agents list   # 登録済みエージェントの一覧

Useful for diagnosing agent spawn failures and Team initialization errors. Check this when the Implementer/Reviewer does not start during Breezing.

Authentication diagnostics (CC 2.1.41+)

CC 2.1.41 で claude auth subcommands have been added:

claude auth status   # 認証状態の確認
claude auth login    # 再ログイン
claude auth logout   # ログアウト

Useful for diagnosing API key issues and authentication expiration.

For VibeCoders

🔍 より詳しい診断が必要な場合

「もっと詳しく診断して」と言われたら:
→ `/debug` コマンドの使用を推奨

/debug コマンドは以下を診断します:
- セッション状態
- MCP サーバー接続
- プラグイン設定
- メモリ使用状況
- エラーログの詳細

💡 使い方:
チャット欄で `/debug` と入力してください

Troubleshooting Coordination Flow

troubleshoot スキルでの診断
    ↓
一般的な問題を自動チェック
    ↓
┌─────────────────────────────────────────┐
│  解決できた?                            │
├─────────────────────────────────────────┤
│  YES → 修正完了                         │
│  NO  → 複雑な問題の可能性               │
└─────────────────────────────────────────┘
    ↓
/debug コマンドを推奨
    ↓
セッション・MCP・設定の詳細診断
    ↓
診断結果をもとに解決策提示

/debug Cases where this is particularly effective

Case Reason
Not resolved after 3 fix attempts Possible environment configuration issue
MCP tools do not respond Diagnosis of the MCP server connection is needed
Malfunction after resuming a session Session state needs to be checked
It works in other projects Project-specific configuration issue

Preventive Advice

To prevent problems before they occur, the following is recommended:

  1. Regular npm run build: Detect problems early
  2. Frequent commits: Easy to roll back when a problem occurs
  3. Updating Plans.md: Make the change history trackable

Notes

  • Auto-fix is attempted up to 3 times
  • Escalate after 3 failures
  • Always confirm before destructive operations
  • Record the fix history in session-log.md

🔧 Utilizing LSP Features

For problem solving, utilize LSP (Language Server Protocol) to accurately identify the cause.

Problem detection via LSP Diagnostics

🔍 LSP 診断実行中...

📊 診断結果:

| ファイル | 行 | メッセージ |
|---------|-----|-----------|
| src/api/user.ts | 23 | 'userId' は undefined の可能性があります |
| src/components/Form.tsx | 45 | 型 'string' を型 'number' に割り当てることはできません |

→ 2件の問題を検出
→ 上記が「動かない」原因の可能性が高い

Cause tracing via Go-to-definition

🔍 原因追跡

エラー: Cannot read property 'name' of undefined

追跡:
1. src/pages/user.tsx:34 - user.name を参照
2. src/hooks/useUser.ts:12 - user を返す ← Go-to-definition
3. src/api/user.ts:8 - API レスポンスが null の可能性

→ API エラー時のハンドリング不足が原因

How to use it for VibeCoders

Symptom LSP Utilization
"It doesn't work" Identify the error location with Diagnostics
"Where is it wrong?" Trace the cause with Go-to-definition
"Since when has it been broken?" Check the impact of changes with Find-references

Details: docs/LSP_INTEGRATION.md