Releases: shinpr/ai-coding-project-boilerplate
Release v1.4.2
Release Notes - npx Setup Feature
🚀 What's New
NPX Support for Quick Project Setup
You can now initialize a new AI-powered project with a single command:
npx github:shinpr/ai-coding-project-boilerplate my-project✨ Features
- Instant Setup: Create a new project without manual cloning
- Interactive Configuration: Guided setup process with language selection
- Automatic Dependencies: Post-setup script handles all npm installations
- Clean Project Structure: Removes unnecessary boilerplate files automatically
📦 Implementation Details
- Added
bin/create-project.jsfor npx execution - Created setup scripts for project initialization
- Updated package.json with bin configuration
- Enhanced README with npx usage instructions
🎯 Benefits
- Faster project initialization
- Reduced setup friction for new users
- Consistent project structure across teams
- Better developer experience
This feature enables developers to quickly bootstrap AI-assisted coding projects with all the necessary configurations and best practices built-in.
Release v1.4.1
What's Changed
- 🔄 Restructured reference management: Removed redundant reference sections from rule files and centralized them in
rules-index.yaml - 🎯 Enhanced rule-advisor: Shifted from "minimal necessary" to "comprehensive quality" approach for better task completion
- 📊 Improved efficiency: Achieved ~18% reduction in rule file sizes for better context window utilization
Key Improvements
- CLAUDE.md basic rules are now always included in rule selection
- Quality assurance rules automatically included for code modifications
- Better handling of tasks like "fix type errors" with appropriate testing rules
Stats
- Japanese rules: -383 lines (-19.2%) / -6,568 characters (-17.2%)
- English rules: -382 lines (-19.1%) / -11,275 characters (-18.3%)
Release v1.4.0
Overview
This release introduces a comprehensive dynamic rule selection system that significantly improves AI agent efficiency and accuracy. The system reduces rule content by 15-18% while maintaining or improving functionality through intelligent rule selection and enhanced agent capabilities.
Major Features
🎯 Dynamic Rule Selection System
- New
rules-index.yamlfiles in bothdocs/rules-ja/anddocs/rules-en/- Provides metadata for each rule file (description, priority, keywords)
- Enables AI agents to select only relevant rules based on task context
- Optimizes context window usage for better performance
🤖 New rule-advisor Subagent
- Intelligent rule selection agent that maximizes AI execution accuracy
- Selects minimal effective ruleset based on task requirements
- Provides structured guidance including:
- Task essence understanding
- Applicable rules selection
- Past failure pattern recognition
- First action recommendations
📝 Rule Content Optimization
- Japanese rules: Reduced by 353 lines (17.7%) / 5,330 characters (13.9%)
- English rules: Reduced by 353 lines (17.7%) / 9,586 characters (15.5%)
- CLAUDE.ja.md: Reduced by 39 lines (22.3%) / 1,040 characters (25.9%)
- CLAUDE.en.md: Reduced by 38 lines (21.8%) / 2,655 characters (32.1%)
- Removed
canonical-phrases.md- content integrated into other files
📚 Enhanced Documentation
- Updated README files with rule index system explanation
- Added detailed descriptions for all 10 subagents
- Documented the dynamic rule selection benefits
- Improved project structure documentation
Technical Improvements
Rule System Enhancements
- Mandatory integration between TodoWrite and rule-advisor
- Automatic rule selection before task execution
- Meta-cognition support through structured task analysis
- Past failure pattern recognition and avoidance
Agent Behavior Controls
- Stricter implementation approval requirements
- Automatic stops for large-scale changes (5+ files)
- Forced rule-advisor execution on repeated errors
- Enhanced progress tracking through TodoWrite integration
Breaking Changes
canonical-phrases.mdhas been removed - content integrated into other rule files- Rule loading now prioritizes the rule index system over loading all files
Migration Guide
- Update language settings:
npm run lang:enornpm run lang:ja - Review new
rules-index.yamlfiles for rule organization - Utilize rule-advisor for optimal rule selection in tasks
- Update any custom scripts that reference
canonical-phrases.md
Commit History
6872f86feat: Release v1.4.0 - Enhanced documentation and rule-advisor integrationec766f5feat: Complete English translation of rule files from Japanese sources29a418efeat: Complete English translation of agent definitions and commands0dc7b63refactor(rules): Rule system evaluation and optimizationeac63earefactor(rules): CLAUDE.ja.md accuracy improvement and rule-advisor integration5001d38refactor: Complete migration to dynamic rule selection system
Statistics
- Total rule content reduction: ~15-18% across all languages
- 10 specialized subagents now available
- 6 core rule files (down from 7)
- Full bilingual support (Japanese/English)
Future Considerations
- Further optimization opportunities exist in consolidating common concepts across files
- Potential for more granular rule selection based on specific task types
- Enhanced integration between agents for complex workflows
This release represents a significant step forward in making AI-assisted development more efficient and accurate through intelligent rule management and enhanced agent capabilities.
Release v1.3.1
Overview
This release enhances the pre-implementation planning process for Claude Code by introducing self-reflection questions that improve metacognitive abilities and reduce implementation errors.
New Features
Enhanced Pre-Implementation Planning
- Task Command Files: Added new command files (
task.md) for both Japanese and English environments in.claude/commands-ja/and.claude/commands-en/directories - Self-Reflection Questions: Updated the implementation workflow to include three critical self-reflection questions that must be addressed before presenting an implementation plan:
- Task Essence & Rules: Identify the core purpose of the task and determine which development rules should guide the approach
- Rule-Based Prioritization: Based on applicable rules, determine what should be tackled first
- Failure Pattern Recognition: Leverage past experience to identify common failures associated with similar tasks
Improvements
Metacognitive Enhancement
- The new self-reflection mechanism encourages deeper thinking before implementation
- Helps prevent common mistakes by forcing consideration of past failure patterns
- Ensures rule compliance from the very beginning of task execution
Bilingual Support
- Both Japanese (
CLAUDE.ja.md) and English (CLAUDE.en.md) documentation have been updated - Command files are available in both languages to support international development teams
Technical Details
- Version bumped from 1.3.0 to 1.3.1
- No breaking changes
- Backward compatible with existing workflows
Benefits
- Reduced Implementation Errors: By thinking through potential pitfalls before coding
- Better Rule Adherence: Explicit identification of applicable rules before starting
- Improved Planning Quality: More thoughtful and comprehensive implementation plans
- Learning from Experience: Systematic consideration of past failures prevents repetition
Migration Guide
No migration required. The changes are additive and enhance the existing workflow without breaking compatibility.
Release v1.3.0
🌐 Multi-language Support Implementation
Overview
This release introduces comprehensive multi-language support for the entire project, enabling seamless switching between Japanese and English across all documentation, templates, and development assets.
Key Features
Dynamic Language Switching System
- Added
scripts/set-language.jsfor language management - Implemented file copying mechanism for language switching
- Created
.claudelangconfiguration file for persistent language state - Added npm scripts:
lang:ja,lang:en,lang:status
Complete File Structure for Both Languages
- CLAUDE.md: Created
CLAUDE.ja.mdandCLAUDE.en.mdvariants - Rule Files: Duplicated entire
docs/rules/asdocs/rules-ja/anddocs/rules-en/ - Sub-agent Guides: Created language-specific guides in
docs/guides/ja/anddocs/guides/en/ - Agent Definitions: Provided
.claude/agents-ja/and.claude/agents-en/ - Commands: Duplicated
.claude/commands/for both languages - Templates: All documentation templates (ADR, Design Doc, PRD, Work Plans) now available in both languages
Dynamic Reference System
- Implemented dynamic path references (
@docs/rules/) that automatically resolve to the active language - Ensured all internal references maintain consistency after language switching
- Preserved architectural patterns and rule references across languages
Technical Details
Language Switching Mechanism
// Core language paths configuration
const LANGUAGE_PATHS = {
claude: { source: (lang) => `CLAUDE.${lang}.md`, target: 'CLAUDE.md' },
rules: { source: (lang) => `docs/rules-${lang}`, target: 'docs/rules' },
guides: { source: (lang) => `docs/guides/${lang}`, target: 'docs/guides/sub-agents.md' },
commands: { source: (lang) => `.claude/commands-${lang}`, target: '.claude/commands' },
agents: { source: (lang) => `.claude/agents-${lang}`, target: '.claude/agents' }
};Supported Languages
- Japanese (
ja) - Default language - English (
en)
Quality Assurance
- Full translation coverage for all documentation
- Consistent terminology across languages
- Preserved technical accuracy in translations
- Maintained formatting and structure integrity
Migration Impact
- Zero Breaking Changes: Existing functionality remains intact
- Backwards Compatible: Projects using previous versions can upgrade seamlessly
- Default Language: Japanese remains the default language (set during
postinstall)
Usage
Switch to English
npm run lang:enSwitch to Japanese
npm run lang:jaCheck Current Language Status
npm run lang:statusBenefits
- Global Accessibility: Project now accessible to both Japanese and English-speaking developers
- Improved Collaboration: Teams can work in their preferred language
- Consistent Experience: All documentation and tooling available in both languages
- Easy Maintenance: Centralized language management system
Future Considerations
- Framework ready for additional language support
- Language preference can be extended to generated code comments
- Potential for automated translation updates
Release v1.2.4
概要
本リリースでは、Claude Codeのサブエージェント(task-executor)およびスラッシュコマンドの実行精度を大幅に改善しました。責務境界の明確化により、各コンポーネントが適切な範囲で動作するようになり、予期しない自動実行や承認スキップなどの問題を解決しています。
主な変更点
🎯 task-executorの実行精度向上
改善内容
- 実装権限の明確化: 呼び出された時点で既に承認済みであることを明示
- 行動原則の具体化: なぜ即座に実装すべきか、なぜ品質チェックを実行しないかを理由付きで記載
- 汎用的な表現: エージェント/オーケストレーターといった実装固有の用語を排除
効果
- task-executorが迷いなく実装に集中できるようになった
- 不要な承認確認による作業の重複を排除
- 責務分離(実装 vs 品質保証)が明確になった
🔧 スラッシュコマンドの責務境界修正
/planコマンド
- 修正前: 実装フェーズ全体の一括承認を取得し、タスク分解まで実行
- 修正後: 作業計画書の作成と計画承認のみに専念
- 効果: 勝手に次フェーズ(実装)に進む問題を解決
/buildコマンド
- 修正前: 分解済みタスクの実装のみ担当
- 修正後: task-decomposerから開始し、タスク分解→実装→品質保証→コミットまでの完全なサイクルを管理
- 追加:
Think carefullyによる品質監視とエラーパターンの注意深い観察
/designコマンド
- 修正: 「作業計画フェーズへの移行」→「次フェーズへの移行」に汎用化
📋 sub-agents-guideの整合性確保
- task-executorへの指示方法を具体的なコード例で明示
- 権限委譲セクションを追加し、自律実行モード開始後の権限を明確化
- checklistProgress機能を削除(task-executorは常に100%完了で報告するため不要)
技術的詳細
削除された機能
checklistProgress: task-executorの構造化レスポンスから削除- 理由: task-executorは単一タスクを完全実行するため、進捗報告は不要
改善された設計原則
- 責務の明確な分離: 各コマンド・エージェントが明確な境界を持つ
- 二重管理の排除: 詳細な実行フローはsub-agents-guideに一元化
- 汎用性の確保: 特定の実装に依存しない表現を使用
影響範囲
.claude/agents/task-executor.md.claude/commands/build.md.claude/commands/plan.md.claude/commands/design.mddocs/sub-agents-guide.md
アップグレード時の注意
本リリースは内部的な改善のため、既存の使用方法に変更はありません。ただし、以下の動作が改善されます:
/planコマンド使用後、自動的に実装が始まることはなくなりました/buildコマンドは作業計画書から自動的にタスク分解を行うようになりました- task-executorは承認確認なしで即座に実装を開始します
今後の展望
今回の改善により、Claude Codeの自律実行がより予測可能で制御しやすくなりました。各コンポーネントの責務が明確になったことで、今後の機能拡張もより安全に行えるようになります。
Release v1.2.3
🎉 新機能
Claude Code カスタムスラッシュコマンド
Claude Codeの作業効率を大幅に向上させる5つのカスタムスラッシュコマンドを追加しました。
利用可能なコマンド
-
/onboard- プロジェクト開発ルールの読み込み- CLAUDE.mdとルールファイルを自動読み込み
- 開発ルールとアーキテクチャの理解を宣言
-
/design- 要件分析から設計書作成まで- requirement-analyzerによる要件分析
- 規模に応じたADR/Design Doc作成
- 設計承認フェーズで明確に終了
-
/plan- 作業計画とタスク分解- 設計書から作業計画書を作成
- 独立したタスクに分解
- タスク分解完了で明確に終了
-
/implement- 自律実行モードでの実装- タスクの順次実行
- 品質チェックとコミット作成
- 全タスク完了まで自動実行
-
/build- 品質チェックとビルド確認- Biome、TypeScript、テストの実行
- エラー時は自動修正
- コミット可能状態の確認
💡 使用例
# プロジェクト開始時
/onboard
# 新機能の設計
/design ユーザー認証機能を追加したい
# 設計承認後の計画
/plan
# タスク分解後の実装
/implement
# 品質確認
/build⚠️ 注意事項
- 各コマンドは定義されたフェーズ内で完結します
- 次のフェーズへの移行は明示的なコマンド実行が必要です
/implementは自律実行モードのため、開始前に計画内容を確認してください
🔧 技術的変更
.claude/commands/ディレクトリ構造の採用- コマンドファイルのFrontmatter形式での説明追加
$ARGUMENTS変数によるパラメータ受け渡し- サブエージェントとの連携強化
Release v1.2.2
🐛 バグ修正
task-executorの自律性向上
- task-executorが不要な確認を繰り返す問題を修正
- 一度呼び出されたら承認済みとして即座に実装を実行するように明確化
- CLAUDE.mdの「調査OK・実装STOP」ルールはtask-executorには適用されないことを明記
タスク重複実行の防止
- 完了済みタスク(全チェックボックスが
[x])の場合は「既に完了」と報告して終了する仕組みを追加 - 完了済みタスクを誤って渡しても実害なく空振りで終了することを明記
サブエージェントへの指示方法の改善
- task-executorへの指示方法を明確化(タスクファイルパスのみ指定)
- 詳細な要件説明が不要であることを明記し、混乱を防止
📝 ドキュメント更新
.claude/agents/task-executor.md: 実装権限セクションを追加docs/sub-agents-guide.md: task-executorへの明確な指示方法を追加
これらの修正により、サブエージェントシステムの実行精度が向上し、同じタスクの重複実行や不要な確認による作業の停滞が解消されます。
Release v1.2.1
🎯 概要
このリリースでは、document-fixer サブエージェントのAI実行精度を最大化するための重要な改善を行いました。整合性検証を最終フェーズに移動し、2段階修正プロセスを導入することで、最終成果物の品質とAI解釈精度の両方を保証します。
🔧 改善
Document Fixer の実行精度向上
2段階修正プロセスの導入
- 第1次修正: レビュー結果に基づく内容改善に特化
- クリティカル → 重要 → 推奨の優先順位で修正
- 論理的矛盾、誤情報、構造的問題を解決
- 第2次修正: AI解釈精度向上のための最適化に特化
- 曖昧表現の明確化
- 参照の具体化
- 構造の最適化
整合性検証の最終フェーズへの移動
- 従来: 並列レビューの一部として実行
- 改善後: すべての修正完了後に単独実行
- 利点: 修正による新たな不整合を確実に検出・修正
AI実行精度向上のための具体的基準
- 必須修正項目:
- 「必要に応じて」「適切に」等の曖昧表現を具体的条件に置換
- 「これ」「それ」等の代名詞を具体的な対象に置換
- 条件分岐をif-then形式に統一
- 用語・表記の完全統一
エラーハンドリングの強化
- 第1次修正と第2次修正で異なる競合解決ルールを適用
- ユーザー判断が必要な事項を明確に記録
- 修正不可能なケースの明確な定義
📈 期待される効果
- 品質保証の向上: 最終フェーズでの整合性検証により、修正による副作用を確実に検出
- AI実行精度の最大化: 曖昧性を排除し、AIが誤解なく文書を解釈・実行可能に
- 予測可能性の向上: 明確な修正基準により、一貫した結果を保証
🔄 後方互換性
この変更は内部処理の改善であり、document-fixerの外部インターフェースに変更はありません。既存の使用方法はすべて維持されます。
📝 技術詳細
変更されたファイル
.claude/agents/document-fixer.md- 実行フローと修正基準を全面的に改善
内部処理の変更点
- レビュー戦略から整合性検証を除外(最終フェーズに移動)
- 修正処理を2段階に分離(内容改善 → AI最適化)
- 統合処理に構造化データ処理と明確な優先順位ルールを追加
- 最終検証にAI実行可能性チェックを追加
🚀 今後の展望
このアップデートにより、AI主導のドキュメント品質管理がより確実かつ予測可能になります。今後も実行精度向上のための改善を継続的に行っていきます。
📋 コミット履歴
5c6627b- feat: enhance document-fixer for maximized AI execution precision
Release v1.2.0
🎉 概要
このリリースでは、新しい document-fixer サブエージェントを導入し、document-reviewer サブエージェントを強化しました。これにより、より強力なドキュメント品質管理機能を提供します。
✨ 新機能
Document Fixer サブエージェント
- レビュー結果に基づいてドキュメントを自動修正する新しい
document-fixerサブエージェントを追加 - 複数のレビュー観点を統合し、フィードバックを集約
- 異なるコンテキストで document-reviewer を並列実行
- 分析から修正まで完全に実施
🔧 改善
Document Reviewer の機能強化
- より構造化されたレビュープロセスで document-reviewer を強化
- 観点ベースのレビュー機能を改善
- ドキュメント修正ワークフローとの統合を改善
ドキュメントの更新
- document-fixer の使用方法を含むサブエージェントガイドを更新
- README のサブエージェントリストに document-fixer を追加
- サブエージェント統合パターンの明確性を向上
📝 変更内容
変更されたファイル
.claude/agents/document-fixer.md- 新しいエージェント定義(初期バージョンから簡略化).claude/agents/document-reviewer.md- レビュー機能を強化README.md- サブエージェントリストに document-fixer を追加docs/sub-agents-guide.md- document-fixer の説明を追加してガイドを拡張
🔄 移行に関する注意
破壊的変更はありません。新しい document-fixer エージェントは追加機能であり、既存のエージェントと並行して動作します。
📋 コミット履歴
7a3e98e- sync: document-fixer/reviewer と sub-agents-guide を更新7bff69a- docs: add document-fixer sub-agent to README7fa901a- fix: update document-fixer to remove operation modes38b111e- Merge pull request #8 from shinpr/sync/document-agents-update