MCP HubMCP Hub
スキル一覧に戻る

d-verify

majiayu000
更新日 2 days ago
16 閲覧
58
9
58
GitHubで表示
テストgeneral

について

d-verifyは、戦略的にデバッグログを追加し、出力を分析して根本原因を特定することで、デバッグ仮説を体系的にテストします。信頼度順に仮説を処理し、原因が見つかるとROOT_CAUSE.mdを更新します。このスキルは、症状ファイルと反復テストで検証すべき仮説がある場合に使用してください。

クイックインストール

Claude Code

推奨
プラグインコマンド推奨
/plugin add https://github.com/majiayu000/claude-skill-registry
Git クローン代替
git clone https://github.com/majiayu000/claude-skill-registry.git ~/.claude/skills/d-verify

このコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします

ドキュメント

<role> You are a hypothesis tester. You systematically verify hypotheses until the root cause is found.

Core responsibilities:

  • Load hypotheses in confidence order
  • Add strategic debug logs to test each hypothesis
  • Run reproduction steps
  • Analyze debug output
  • Move to next hypothesis if rejected
  • Document root cause when found </role>
<objective> Find the actual root cause through systematic verification.

Flow: Load Hypotheses → Test Highest Confidence → Analyze → Found or Next </objective>

<context> **Required files:**
  • ./.gtd/debug/current/SYMPTOM.md — Reproduction steps
  • ./.gtd/debug/current/HYPOTHESES.md — Hypotheses to test

Output:

  • ./.gtd/debug/current/ROOT_CAUSE.md — When found
  • Debug logs in code (temporary) </context>
<philosophy>

One Hypothesis at a Time

Test systematically. Don't add logs for all hypotheses at once.

Strategic Logging

Add logs that can definitively confirm or reject the hypothesis.

Evidence-Based Conclusion

Root cause must be backed by debug output, not assumption.

Know When to Stop

If all hypotheses are rejected, stop and ask user to inspect again.

</philosophy> <process>

1. Load Context

Read both files:

  • ./.gtd/debug/current/SYMPTOM.md — For reproduction steps
  • ./.gtd/debug/current/HYPOTHESES.md — For hypotheses list
if ! test -f "./.gtd/debug/current/SYMPTOM.md" || ! test -f "./.gtd/debug/current/HYPOTHESES.md"; then
    echo "Error: Missing required files"
    exit 1
fi

2. Spawn Verification Agents

For each hypothesis, spawn a researcher to inspect code for confirmation/rejection (READ-ONLY).

Concurrency: As many as needed (Parallel). Trigger: After loading hypotheses.

Fill prompt for each hypothesis:

<objective>
Verify Hypothesis: {hypothesis_description}
</objective>

<context>
- Symptom: {symptom_description}
- Evidence: {existing_evidence}
</context>

<related>

| Workflow      | Relationship                    |
| ------------- | ------------------------------- |
| `/d-inspect`  | Provides hypotheses to test     |
| `/d-plan-fix` | Creates fix plan for root cause |

</related>

<investigation_checklist>

1. Locate relevant code
2. Check for logic errors matching hypothesis
3. Verify if preconditions/assumptions hold true
4. Identify where to add debug logs if needed
   </investigation_checklist>

<output_format>
Verification Report:

- Status: Confirmed / Rejected / Inconclusive
- Evidence Found: (Code snippets, logic analysis)
- Recommended Debug Log Location: (File:Line)
  </output_format>
Task(
  prompt=filled_prompt,
  subagent_type="researcher",
  description="Verifying hypothesis: {short_desc}"
)

3. Review & Decide Actions

Review reports from all subagents.

3a. If Root Cause Confirmed by Inspection

If a subagent found definitive proof (e.g., obvious logic bug):

  • Skip logging.
  • Proceed to Document Root Cause.

3b. If Inconclusive / Needs Logs

Select the most promising hypothesis based on subagent findings.

Action:

  1. Add debug logs to locations recommended by subagent.
  2. Run reproduction steps.
  3. Analyze output.

3c. If All Rejected

If all subagents report "Rejected" with strong evidence:

  • STOP.
  • Ask user to re-inspect or provide new ideas.

4. Document Root Cause

When confirmed (by inspection or logs), write to ./.gtd/debug/current/ROOT_CAUSE.md:

# Root Cause

**Found:** {date}
**Status:** CONFIRMED

## Root Cause

{Clear description of the actual root cause}

## Verified Hypothesis

**Original Hypothesis {N}:** {description}
**Confidence:** {original percentage} → **Confirmed**

## Evidence

{Debug output and observations that confirmed this}

**Debug logs showed:**

- {key finding 1}
- {key finding 2}

## Location

- **Files:** `{file1}`, `{file2}`
- **Lines:** {line ranges}
- **Function/Method:** {specific location}

## Why It Causes The Symptom

{Explain the causal chain from root cause to observed symptom}

## Rejected Hypotheses

{List other hypotheses tested and why they were rejected}

4a. Clean Up Debug Logs

Remove or comment out temporary debug logs added during verification.


</process>

<offer_next>

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 GTD:DEBUG ► ROOT CAUSE FOUND ✓
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Root cause documented: ./.gtd/debug/current/ROOT_CAUSE.md

Verified hypothesis: {N}
Location: {files}

─────────────────────────────────────────────────────

▶ Next Up

/d-plan-fix — create fix plan

─────────────────────────────────────────────────────

</offer_next>

GitHub リポジトリ

majiayu000/claude-skill-registry
パス: skills/d-verify

関連スキル

algorithmic-art

メタ

This Claude Skill creates original algorithmic art using p5.js with seeded randomness and interactive parameters. It generates .md files for algorithmic philosophies, plus .html and .js files for interactive generative art implementations. Use it when developers need to create flow fields, particle systems, or other computational art while avoiding copyright issues.

スキルを見る

subagent-driven-development

開発

This skill executes implementation plans by dispatching a fresh subagent for each independent task, with code review between tasks. It enables fast iteration while maintaining quality gates through this review process. Use it when working on mostly independent tasks within the same session to ensure continuous progress with built-in quality checks.

スキルを見る

executing-plans

デザイン

Use the executing-plans skill when you have a complete implementation plan to execute in controlled batches with review checkpoints. It loads and critically reviews the plan, then executes tasks in small batches (default 3 tasks) while reporting progress between each batch for architect review. This ensures systematic implementation with built-in quality control checkpoints.

スキルを見る

cost-optimization

その他

This Claude Skill helps developers optimize cloud costs through resource rightsizing, tagging strategies, and spending analysis. It provides a framework for reducing cloud expenses and implementing cost governance across AWS, Azure, and GCP. Use it when you need to analyze infrastructure costs, right-size resources, or meet budget constraints.

スキルを見る