Release v1.11.2

What's New

This release prepares the package for npm publication with a new, shorter package name.

Changes

• New package name: create-ai-project (previously ai-coding-project-boilerplate)
• npm init support: Now supports npm init ai-project shorthand

Installation

npx create-ai-project my-project
# or
npm init ai-project my-project

Options

npx create-ai-project my-project --lang=ja  # Japanese
npx create-ai-project my-project --lang=en  # English (default)

Release v1.11.1

Shallow Thinking Suppression for Technical Design Documents

Added elements to ADR and Design Doc templates that force explicit acknowledgment of uncertainties and failure scenarios, reducing LLM overconfidence.

Changes

ADR Template

• Added "Decision Details" section with Why now / Why this / Known unknowns / Kill criteria

Design Doc Template

• Added "Design Summary (Meta)" block with risk level, constraints, and unknowns

document-reviewer Agent

• Added "Failure scenario review" perspective covering normal usage, high load, and external failures

Release v1.11.0

New Features

New Sub-Agents

• integration-test-reviewer: Verifies test implementation quality by checking consistency between skeleton comments and actual test code
• design-sync: Ensures consistency across Design Documents

Testing Improvements

• EARS notation for acceptance criteria (Event-Action-Response-State format)
• Property-based testing support with fast-check integration
• Property annotations in test skeletons for invariant verification

Quality Check Reorganization

• New check:code script consolidates code quality checks (lint, unused exports, circular deps, build)
• Streamlined 5-phase quality check workflow

Improvements

• Clarified quality-fixer phase structure (removed ambiguous Phase 6)
• Enhanced acceptance-test-generator with ROI-based test selection
• Improved sub-agent documentation clarity

Full Changelog

v1.10.0...v1.11.0

Release v1.10.0

Package Manager Flexibility

You can now switch between npm, bun, or pnpm by updating package.json:

{
  "packageManager": "[email protected]",
  "scripts": {
    "build": "bun run tsc && tsc-alias",
    "test": "bun test"
  }
}

Claude Code will automatically use the appropriate commands based on your configuration.

See the new Package Manager Configuration section for details.

Release v1.9.3

Changes

• Remove ambiguous "Proactive Use of Specialized Agents" section from CLAUDE.md
  • rule-advisor guidance was redundant (already in meta-cognition section)
  • quality-fixer trigger conditions were unclear and imprecise
• Reduces CLAUDE.md token count for improved clarity

Release v1.9.2

Improvements

Rule Advisor Optimization

• Enhanced TodoWrite integration with rule-advisor output
  • firstActionGuidance added as first Todo item
  • taskEssence recorded as completion criteria
  • warningPatterns tracked as execution checkpoints

Pink Elephant Problem Fix

Replaced prohibition-focused expressions with positive action alternatives to improve LLM execution accuracy.

Before	After
"Prohibited Actions"	"Required Alternative Patterns"
"any type is prohibited"	"use unknown type with type guards"
"Prohibited Fix Patterns"	"Correct Fix Patterns"

Files Changed

• CLAUDE.md (ja/en)
• coding-standards.md (ja/en)
• quality-fixer.md (ja/en)
• quality-fixer-frontend.md (ja/en)
• task-decomposer.md (ja/en)

Release v1.9.1

🔧 Improvements

Rule Organization

• Simplified Fail-Fast principle documentation to eliminate duplication and improve maintainability
  • Reduced from 16 lines to 4 lines in coding-standards.md
  • Removed implementation details, keeping only core principles
  • Implementation specifics now live in language-specific rules (typescript.md, frontend/typescript.md)

📝 Changes

• Refactored docs/rules-*/coding-standards.md - Simplified error handling section
• Updated docs/rules-*/rules-index.yaml - Section name changes
• Bumped version to 1.9.1

🎯 Benefits

• Single source of truth: Core principle defined once, implementation details in appropriate files
• Reduced maintenance burden: Updates only needed in language-specific rules
• Better separation of concerns: Universal principles vs. implementation patterns

---

Full Changelog: v1.9.0...v1.9.1

Release v1.9.0

Overview

This release focuses on reducing AI context pollution, eliminating documentation duplication, and implementing intelligent test generation strategies. Major improvements include ROI-based test filtering, universal coding standards consolidation, and enhanced agent prompt precision.

🎯 Highlights

• 60-75% Test Reduction: ROI-based filtering prevents over-generation while maintaining quality
• ~760 Lines Removed: Eliminated Frontend/Backend rule duplication
• Future-Proof Searches: Dynamic year detection for latest information queries
• 130 Lines Net Reduction: Agent prompt optimization reduces context pollution

🚀 Features

ROI-Based Test Generation (#91)

Implement 3-layer quality filtering to prevent test over-generation:

• Behavior-First Filtering: Remove implementation details at AC level
• Two-Pass Generation: Enumerate candidates → ROI-based selection
• Budget Enforcement: Hard limits (max 3 integration, 1-2 E2E per feature)

Impact: Expected 60-75% test reduction with 3-5x ROI improvement

Technical Details:

ROI Score = (Business Value × User Frequency + Legal Requirement × 10 + Defect Detection)
          / (Creation Cost + Execution Cost + Maintenance Cost)

Cost Table:

• Unit: 3 (create: 1, execute: 1, maintain: 1)
• Integration: 11 (create: 3, execute: 5, maintain: 3)
• E2E: 38 (create: 10, execute: 20, maintain: 8)

Universal Coding Standards (#90)

Consolidate 90% duplicated content from Frontend/Backend documentation:

• Created: coding-standards.md with universal coding principles
• Refactored: Backend/Frontend-specific rules separated
• Removed: Redundant ai-development-guide.md files
• Updated: 15 agent files to reference new structure

Impact:

• Before: ~90% overlap between Frontend/Backend docs
• After: Single source of truth + domain-specific rules
• Result: 33 files changed, +429/-1,188 lines (~760 net reduction)

🔧 Improvements

Agent Prompt Optimization

Remove context pollution and improve LLM precision:

• Removed skip tracking from acceptance-test-generator (no longer needed)
• Simplified AC scoping in technical-designer (removed redundant 3-check process)
• Dynamic year fetching for latest information research (date +%Y)
• Eliminated hardcoded years (2024/2025) from search patterns

Impact: +288/-418 lines (130 net reduction), improved token efficiency

Latest Information Research

Future-proof search queries with dynamic year detection:

# Before
`React Server Components best practices 2024`

# After
date +%Y  # e.g., 2025
`React Server Components best practices {current_year}`

Applied to:

• technical-designer.md (en/ja)
• document-reviewer.md (en/ja)

🐛 Bug Fixes

Initialization Script Logic (#89)

• Fix template file detection to use filename instead of relative path
• Remove redundant set-language.js copy process
• Add setup-project.js to cleanup list for user projects

📊 Metrics

Category	Before	After	Change
Test Generation	100% of candidates	25-40% (top ROI)	-60-75%
Documentation	~1,950 lines (duplicated)	~1,190 lines (consolidated)	-760 lines
Agent Prompts	418 lines (skip tracking)	288 lines (optimized)	-130 lines
Search Accuracy	Hardcoded years (2024)	Dynamic year fetch	Future-proof

📝 Migration Guide

For Existing Projects

1. Test Generation: No migration needed. New ROI-based filtering applies automatically.

2. Coding Standards: Update rule file references if customized:

   - @docs/rules/ai-development-guide.md
   + @docs/rules/coding-standards.md

3. Latest Information Searches: Agents will now dynamically fetch current year. No manual updates needed.

Release: v1.8.0

🎉 What's New

Frontend Development Infrastructure

Added comprehensive support for React-based web application development alongside existing backend capabilities.

New Features:

• 3 Specialized Frontend Agents

  • technical-designer-frontend: Component architecture and design decisions
  • task-executor-frontend: TDD-driven implementation with React Testing Library
  • quality-fixer-frontend: Quality checks (types, tests, coverage, bundle size)

• 3 Frontend Slash Commands

  • /front-design: Requirements → Design Doc
  • /front-plan: Design Doc → Work plan
  • /front-build: Autonomous task execution

• Frontend-Specific Rules

  • Props-driven component design
  • React Testing Library + MSW patterns
  • Type-safe data flow (React state management)
  • 60%+ test coverage with component-specific targets

Bilingual Support:

• Full EN/JA documentation for all agents, commands, and rules

Technology-Agnostic Approach

Removed version-specific dependencies to improve portability:

• React 19 → React (version-agnostic)
• Vite → Build tool (tool-agnostic)
• Removed Lighthouse-specific mentions

Other Improvements

• CI Workflow: Updated to work without package-lock.json
• Props Guidelines: 3-7 props ideal, 50% optional max, 2-level nesting
• Coverage Targets: Component-specific (Atoms 70%, Molecules 65%, Organisms 60%)

📊 Stats

• 30 files changed: +4,361 / -6,864 lines
• New documentation: 2,100+ lines of frontend-specific rules
• New agents: 1,900+ lines of specialized configurations

---

Full Changelog: v1.7.17...v1.8.0

Release v1.7.17

Documentation Improvements

Clarified rule-advisor Agent Invocation Method

Problem: Claude Code was incorrectly attempting to invoke the rule-advisor agent using Python or other tools instead of the Task tool, leading to execution failures.

Solution: Added explicit clarification in the execution prerequisites section (both EN and JA documentation) that rule-advisor agent must be invoked using the Task tool.

Changed Files:

• CLAUDE.en.md: Updated execution prerequisites to specify "rule-advisor agent execution (invoke using Task tool)"
• CLAUDE.ja.md: Updated execution prerequisites to specify "rule-advisorエージェント実行(Taskツールで呼び出すこと)"

Impact: This minimal annotation ensures consistent and correct agent invocation across all usage contexts.