Nabledge-6: Nablarch 6 Knowledge Base

Structured knowledge base for Nablarch 6 framework, covering batch processing and RESTful web services.

What this skill provides

Knowledge Coverage: Batch, REST, handlers, libraries (UniversalDao, DB access, validation, file I/O, business date), testing (NTF), adapters, security

Code Analysis: Dependency tracing, component decomposition, architecture visualization, documentation generation

Use cases:

• Knowledge search: Learn Nablarch APIs, patterns, configurations, troubleshooting
• Code analysis: Understand existing code structure, dependencies, and design
• Development support: Feature implementation, code review, error investigation

How to use

Basic usage

Interactive mode (recommended):

nabledge-6

Shows friendly greeting and lets you choose between knowledge search and code analysis.

Direct knowledge search:

nabledge-6 "Your question about Nablarch"

Direct code analysis:

nabledge-6 code-analysis

Important constraint: Knowledge files only

CRITICAL: Always answer using ONLY the information in knowledge files (knowledge/*.json).

• DO NOT use LLM training data or general knowledge about Nablarch
• DO NOT access official websites or external resources during answers
• DO NOT guess or infer information not present in knowledge files
• If knowledge is missing: Clearly state “この情報は知識ファイルに含まれていません” and list related available knowledge

Why this constraint: Knowledge files contain verified, accurate information extracted from official documentation. LLM training data may be outdated or incomplete.

Error Handling Policy

General principle:

• Always inform user clearly when something goes wrong
• Provide actionable next steps
• Never fail silently

Knowledge not found (Knowledge Search):

• Message: “この情報は知識ファイルに含まれていません”
• List related available knowledge from index.toon
• Show “not yet created” entries if applicable
• DO NOT use LLM training data or general knowledge

Target code not found (Code Analysis):

• Message: “指定されたコードが見つかりませんでした”
• Show search patterns used (e.g., **/*LoginAction.java)
• Ask for clarification (more specific file name, module, or full path)

Workflow execution failure:

• Inform user which step failed (e.g., “Step 2: 依存関係分析中にエラーが発生しました”)
• Show error details if available
• Suggest retry or alternative approach

Output file already exists (Code Analysis):

• Ask user: “上書きする” / “別名で保存” / “キャンセル”

Dependency analysis too complex (Code Analysis):

• Ask user to narrow scope
• Suggest limiting to direct dependencies only
• Provide partial analysis with note

How Claude Code should execute this skill

CRITICAL: When this skill is invoked, Claude Code MUST execute workflows manually. The skill does NOT automatically process – Claude must follow the workflow steps using tools.

Execution Process

When you (Claude Code) receive this skill prompt, follow these steps:

Step 0: Check arguments and determine workflow

Decision tree:

1. No arguments (nabledge-6):

   • Proceed to Step 1: Show greeting and ask user choice

2. Argument: “code-analysis” (nabledge-6 code-analysis):

   • Skip to Code Analysis Workflow (Step 3)

3. Other text arguments (nabledge-6 <question>):

   • Treat argument as user question
   • Skip to Knowledge Search Workflow (Step 2)

Step 1: User-friendly greeting (no arguments case only)

Tool: AskUserQuestion

Action: Show friendly greeting and ask what user wants to do

Message:

Nablarch 6のことなら何でもお答えします。

以下のようなことが可能です:
- Nablarchの機能や使い方について質問する
  例: UniversalDaoの使い方、バッチ処理の実装方法、トランザクション管理
- 既存コードの構造を理解するためのドキュメントを生成する
  例: LoginActionの構造、プロジェクト全体のアーキテクチャ

何をお手伝いしましょうか？

Options:

• Option 1: “Nablarchの機能や使い方を知りたい” → Proceed to Step 2 (Knowledge Search)
• Option 2: “既存コードの構造を理解したい” → Proceed to Step 3 (Code Analysis)
• Other (free text input) → Treat as knowledge search question, proceed to Step 2

Step 2: Knowledge Search Workflow

Execute when user wants to search Nablarch knowledge.

Inform user: “Nablarch知識ベースを検索します”

Execute workflows:

Follow the workflows in this order:

1. keyword-search workflow (see workflows/keyword-search.md):

   • Read knowledge/index.toon
   • Extract keywords (3 levels: domain, component, functional)
   • Match against index hints
   • Select top 10-15 files
   • Extract section indexes using jq
   • Build candidates list (20-30 sections)

2. section-judgement workflow (see workflows/section-judgement.md):

   • Read each candidate section
   • Judge relevance: High (2), Partial (1), None (0)
   • Filter out None relevance
   • Sort by relevance (High first)
   • Return top 10-15 sections (~5,000 tokens)

3. Answer using knowledge files only:

   • Extract information from High and Partial sections
   • Format as clear, structured answer
   • ONLY use information from knowledge files
   • Cite sources (e.g., “universal-dao.json:paging section”)
   • DO NOT supplement with LLM training data
   • Priority order: Accuracy > Brevity > Completeness (if knowledge files lack information for full structure, provide shorter answer rather than supplementing)

   Output format constraints (to keep answers concise):

   • Target length: 500 tokens or less (strict limit for simple queries; may extend to 800 tokens for multi-part questions)
   • If answer naturally exceeds limit: Prioritize 結論 section and provide knowledge file references for details
   • Required structure:
     • 結論 (Conclusion): Direct answer to user’s question
     • 根拠 (Evidence): 1 code example from knowledge files
     • 注意点 (Considerations): Important points or limitations
   • For detailed/complex topics: Provide summary + reference to knowledge file path
     • Example: “詳しくは knowledge/features/libraries/universal-dao.json#paging を参照”
   • Use Japanese section names (結論、根拠、注意点) since users are Japanese

4. Handle missing knowledge (if needed):

   • State clearly: “この情報は知識ファイルに含まれていません”
   • List related available knowledge from index.toon
   • DO NOT answer from LLM training data

Tools: Read, Grep (optional), Bash with jq Expected: ~10-15 tool calls, ~5,000 tokens

Step 3: Code Analysis Workflow

Execute when user wants to analyze existing code.

Entry conditions:

• User selected Option 2 from Step 1
• Skill invoked with nabledge-6 code-analysis

Inform user: “既存コードを分析してドキュメントを生成します”

If target code not specified: Ask user for target specification using AskUserQuestion

Execute workflow:

Follow workflows/code-analysis.md:

1. Identify target and analyze dependencies:

   • Parse user request (class/feature/package)
   • Find target files using Glob/Grep
   • Read target files and extract dependencies
   • Classify dependencies (project/Nablarch/libraries)
   • Build dependency graph
   • Categorize components by role

2. Search Nablarch knowledge:

   • For each Nablarch component, execute keyword-search workflow
   • Collect relevant knowledge sections

3. Generate and output documentation:

   • Read template files (MUST READ FIRST)
   • Build Mermaid classDiagram and sequenceDiagram
   • Create component summary table
   • Write component details with line references
   • Write Nablarch usage with important points (✅ ⚠️ 💡)
   • Apply template with all placeholders
   • Calculate analysis duration
   • Write file to .nabledge/YYYYMMDD/code-analysis-<target>.md
   • Inform user of completion

Tools: Read, Glob, Grep, Bash with jq, Write Expected: ~30-50 tool calls, 1 documentation file

Workflow Files

• workflows/keyword-search.md: Keyword-based knowledge search (3 steps)
• workflows/section-judgement.md: Relevance judgement for knowledge sections (2 steps)
• workflows/code-analysis.md: Existing code analysis and documentation generation (3 steps)

Template Files

• assets/code-analysis-template.md: Documentation template for code analysis output
• assets/code-analysis-template-guide.md: Template usage guide (sections, placeholders, evaluation criteria)
• assets/code-analysis-template-examples.md: Template examples (component table, Nablarch usage, links)

Quick reference

Interactive mode:

nabledge-6

Shows greeting → Choose knowledge search or code analysis

Knowledge search examples:

nabledge-6 "Nablarchでページングを実装したい"
nabledge-6 "UniversalDaoの使い方"
nabledge-6 "トランザクション管理ハンドラのエラー対処"

Process: keyword-search → section-judgement → answer using knowledge files only

Code analysis examples:

nabledge-6 code-analysis

Then specify: “LoginActionを理解したい”, “proman-batchモジュール全体の構造“, etc. Process: Identify → Analyze → Search knowledge → Generate docs → Output file

Knowledge structure

Knowledge files (JSON format):

• knowledge/features/: Handlers, libraries, processing methods, tools, adapters
• knowledge/checks/: Security checklist, public API list, deprecated features
• knowledge/releases/: Release notes

Index (TOON format):

• knowledge/index.toon: 93 entries with ~650 search hints

Human-readable (auto-generated):

• docs/: Markdown version of knowledge files for human verification

Each knowledge file includes:

• official_doc_urls: Source URLs for verification
• index: Section-level search hints
• sections: Structured knowledge by topic

Example: knowledge/features/libraries/universal-dao.json

Advanced usage

Manual search:

grep -i "ページング" knowledge/index.toon

Read specific knowledge:

jq '.sections.paging' knowledge/features/libraries/universal-dao.json

Browse human-readable version:

cat docs/features/libraries/universal-dao.md

Token efficiency

Index: ~5,000-7,000 tokens (TOON format, 40-50% reduction vs JSON) Search results: ~5,000 tokens (top 10 sections, 500 tokens each) Total: ~10,000-12,000 tokens (5-6% of 200k context window)

Version information

Target version: Nablarch 6u2 / 6u3

Out of scope:

• Jakarta Batch
• Resident batch (table queue)
• Web applications (JSP/UI)
• Messaging (MOM)

Quality assurance

Knowledge accuracy: Average 97.3/100 points (verified against official documentation)

Coverage: 17 files created, 43 files planned (total 60 files)

Source: Official documentation (https://nablarch.github.io/docs/), Fintan system development guide

Limitations

Knowledge coverage

Not yet created knowledge files show “not yet created” in index.toon (76 out of 93 entries).

When knowledge is missing:

1. Clearly state: “この情報は知識ファイルに含まれていません”
2. List related available knowledge that might help
3. Show the entry from index.toon with “not yet created” status
4. DO NOT attempt to answer from LLM training data or general knowledge
5. ONLY IF EXPLICITLY REQUESTED: Provide official_doc_urls for manual reference

Current coverage (17 files):

• Nablarch batch processing basics
• Core handlers (DB connection, transaction, data read)
• Core libraries (UniversalDao, database access, file path, business date, data bind)
• Testing framework (NTF) basics
• SLF4J adapter
• Security checklist
• Release notes (6u3)

Verification

Human verification: Check docs/ directory for human-readable versions. All knowledge includes official_doc_urls showing the source.

Accuracy: Average 97.3/100 points (verified against official RST documentation)

Feedback

If knowledge is inaccurate or missing, please:

1. Check official_doc_urls in the knowledge file for the source
2. Verify against official documentation
3. Report discrepancies to the knowledge maintainer

References (for manual lookup only)

IMPORTANT: These references are for human users to manually look up information NOT in knowledge files. Do not access or fetch these during answers.

• Nablarch Official Documentation – Use when knowledge file says “not yet created”
• Fintan System Development Guide – Patterns and anti-patterns
• Nablarch Example Batch – Code examples
• Nablarch Example REST – REST examples