CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.18.5] - 2026-03-29

Fixed

Dependency Existence Verification in Design Workflow

Design documents could describe components as "existing" without verification, causing implementation failures when those dependencies turned out to be missing or mismatched.

• technical-designer / technical-designer-frontend: Add Dependency Existence Verification step to Existing Code Investigation — each assumed dependency is now verified via Grep/Glob before design proceeds, with three-way classification (found in codebase / external dependency / requires new creation)
• document-reviewer: Add dependency realizability check to Gate 1 — cross-checks claimed "existing" dependencies against the codebase during review. Missing dependency with no external source → critical (feasibility). Signature mismatch (method names, parameter types, return types) → important (consistency)

[1.18.4] - 2026-03-28

Fixed

Work-Planner Invocation Clarity

• subagents-orchestration-guide: Small Scale flow now explicitly names work-planner agent
• plan command: Step 3 rewritten with explicit subagent_type and conditional prompt branching (with/without test skeletons)
• front-plan command: Differentiate placeholder names ([integration test path] / [E2E test path] instead of identical [path from step 2])
• build / front-build commands: Split "Neither exists" state into Design Doc present (invoke work-planner) vs absent (report and stop)
• implement command: Add explicit subagent_type: "work-planner" to test information communication section

Skills Index Sync

• skills-index.yaml (en): Remove stale parenthetical from coding-standards Security Principles section name
• skills-index.yaml (en): Add missing Test Design Patterns section to frontend-typescript-testing
• skills-index.yaml (ja): Remove stale References and アンチパターン sections from frontend-typescript-testing, add missing テスト設計パターン

[1.18.3] - 2026-03-26

Changed

Test Workflow: Skeleton-Based Single Source of Truth

• design-template: Remove Test Strategy section — test design responsibility moved to acceptance-test-generator skeletons
• design-template: Integration Points change from Verification to Contract (interface/API contract)
• plan-template: Remove Operational Verification Procedures from all phases — replaced by test skeleton file path references
• plan-template: Completion criteria updated to reference skeleton-based test results
• work-planner: Replace operational verification with test skeleton placement rules (integration tests in corresponding phases, E2E in final phase)
• work-planner: Add fallback for when test skeletons are not provided (AC-based test planning)
• documentation-criteria: Design Doc definition removes E2E verification; Work Plan references skeleton file paths
• task-decomposer: Phase completion content references skeleton file paths instead of copying E2E verification from Design Doc
• technical-designer / technical-designer-frontend: Replace E2E verification checklist with testable AC quality gate (concrete trigger, action, expected result)
• technical-designer-frontend: Add typescript-testing skill with Applying to Implementation entry

add-integration-tests: Fullstack Support

• Support backend, frontend, and fullstack test generation (previously backend-only)
• Document discovery with filename-based classification and user confirmation GATE
• Per-Design-Doc skeleton generation (aligned with acceptance-test-generator single-doc contract)
• Layer-specific task file naming for deterministic agent routing (*-backend-task-* / *-frontend-task-*)
• Input validation for empty arguments

[1.18.2] - 2026-03-26

Changed

Investigation Depth Consistency

• code-verifier: Section-by-section claim extraction with minimum 20 claims threshold and score stability constraint
• code-verifier: Tool-backed evidence rules requiring Glob/Grep confirmation for existence, behavioral, and identifier claims
• code-verifier: Reverse coverage assessment (code-to-document direction) enumerating routes, test files, and public exports
• investigator: Structured source table with minimum investigation actions per source type, replacing vague bullet list
• investigator: Tracking depth check replacing "shallow tracking signs" with explicit stop condition requirement
• verifier: Structured triangulation supplementation with source type gap analysis

Reverse-Engineer Workflow Accuracy

• prd-creator: 7-step reverse investigation protocol (route enumeration → entry point tracing → data model → tests → roles → spec → confirmation)
• prd-creator: Exact code transcription requirement for identifiers, URLs, and parameter names
• prd-creator: Scope expansion from entry points when external scope is provided
• scope-discoverer: Unit inventory enumeration (routes, test files, public exports) per discovered unit, propagated to downstream agents
• scope-discoverer: Step renumbering (4.5 → 5, subsequent steps shifted to 6-9)
• technical-designer / technical-designer-frontend: Reverse-engineer operation mode with mode-separated checklists
• reverse-engineer: Independent verifier scope discovery (code_paths intentionally omitted)
• reverse-engineer: verifiableClaimCount ≥ 20 quality gate
• reverse-engineer: unitInventory propagation to Design Doc generation

Diagnostic Workflow

• diagnose: Structured investigator prompts passing taskEssence and investigationFocus
• diagnose: Expanded quality checks with 6 specific criteria and re-run prompt template

Agent Structure

• Removed redundant "Core Responsibilities" sections from 5 agents (code-verifier, investigator, scope-discoverer, verifier) — content preserved in execution steps
• technical-designer / technical-designer-frontend: Consolidated integration point contracts, removed ADR template duplication
• technical-designer / technical-designer-frontend: Research guidelines condensed with reverse-engineer mode skip rule

[1.18.1] - 2026-03-25

Added

Structured JSON Output Enforcement

• Explicit "Return JSON Result" step added to all 15 agents (EN/JA) to ensure consistent structured JSON output
• Completion criteria updated with "Final response is the JSON output" check across all agents

Scope Discovery Improvements

• scope-discoverer: PRD unit grouping (step 7) with valueProfile metadata (targetPersona, userGoal, valueCategory)
• scope-discoverer: prdUnits array in output format for downstream document generation
• reverse-engineer: Updated to use prdUnits with quality gate validation (sourceUnits completeness check)
• reverse-engineer: Human review point now presents PRD unit grouping for confirmation

Document Update Layer Detection

• update-doc: Auto-detect frontend/backend layer from document content for correct agent selection (technical-designer vs technical-designer-frontend)

Changed

• quality-fixer / quality-fixer-frontend: "User Report" section replaced with "Intermediate Progress Report" to clarify final output must be JSON
• scope-discoverer: "Merge signals" renamed to "Cohesion signals" with note that grouping is deferred to step 7
• investigator: Step 4 renamed from "Impact Scope Identification and Output" to "Impact Scope Identification"
• solver: Step 5 renamed from "Implementation Steps Creation and Output" to "Implementation Steps Creation"

Fixed

• Added missing "Applying to Implementation" sections to 5 Japanese agent definitions (code-reviewer, acceptance-test-generator, design-sync, integration-test-reviewer, requirement-analyzer) that caused JA agents to operate without skill loading instructions

[1.18.0] - 2026-03-21

Added

Security Review Integration

• New security-reviewer agent (EN/JA) with 4-category finding classification (confirmed_risk, defense_gap, hardening, policy) and structured rationale
• Security check reference (references/security-checks.md) with stable and trend-sensitive detection patterns in coding-standards skill
• Security Principles section (Secure Defaults, Input and Output Boundaries, Access Control, Knowledge Cutoff Supplement) in coding-standards skill
• Security Review step in build, front-build, implement commands — invoked after all task cycles complete
• security-reviewer integrated into review and front-review commands as parallel validation alongside code-reviewer
• requiresTestReview boolean field in task-executor/task-executor-frontend structured responses
• Prompt Construction Rule in subagents-orchestration-guide for consistent subagent invocation
• Security considerations section in design-template with concrete evaluation items (Authentication, Input Validation, Sensitive Data)
• Security review task in plan-template final quality assurance phase

Changed

Agent Pipeline Improvements

• code-reviewer: Rewrite from checklist to 5-step verification pipeline; add acceptanceCriteria array with per-AC status/location/gap/suggestion; change Required Information to Input Parameters
• requirement-analyzer: Rewrite from responsibility list to 6-step verification process; integrate file count estimation into Step 2; change Required Information to Input Parameters
• work-planner: Rewrite from responsibility list to 6-step planning process; change Required Information to Input Parameters
• subagents-orchestration-guide: Add security-reviewer to agent list and response specs; update flow diagram; change testsAdded pattern matching to requiresTestReview boolean
• task-analyzer/skills-index.yaml: Add Security Principles detail to coding-standards sections

[1.17.1] - 2026-03-19

Fixed

• Fix incorrect tool name: "Task tool" → "Agent tool" in all command files (EN/JA)
• Update deprecated MSW v1 API (rest/res/ctx) to v2 (http/HttpResponse) in frontend-typescript-testing skill
• Update deprecated fireEvent to userEvent.setup() pattern per RTL best practices
• Convert "Prohibited Actions" to "Output Self-Check" checklists in 6 agents for more effective LLM guidance
• Clarify orchestrator delegation protocol with explicit Agent tool reference and permitted tools whitelist
• Add sequential execution ordering and as-is data bridging rules to reverse-engineer orchestrator protocol
• Remove internal implementation details from skill descriptions (skill-optimization, documentation-criteria)
• Reorder test design pattern examples good-first to avoid LLM pattern imprinting

Added

• affectedLayers field in requirement-analyzer output for fullstack workflow support
• Test design patterns section (good/bad examples) in EN frontend-typescript-testing skill (previously JA-only)

[1.17.0] - 2026-03-08

Added

UI Spec Support

• New ui-spec-designer agent (EN/JA) for creating UI Specifications from PRD and optional prototype code
• UI Spec template in documentation-criteria references
• UI Spec document type recognized by document-reviewer
• UI Spec integration steps in technical-designer-frontend
• UI Spec phase added to front-design command and orchestration flows

Playwright E2E Testing

• E2E test implementation guide (references/e2e.md) in frontend-typescript-testing skill
• E2E test design guide (references/e2e-design.md) in integration-e2e-testing skill
• E2E test support in quality-fixer-frontend and acceptance-test-generator

Autonomous Execution Enhancements

• 4-step task execution cycle (task-executor → escalation check → quality-fixer → commit) in build and front-build commands
• integration-test-reviewer integration for test skeleton compliance validation
• Structured response specs and sub-agent invocation constraints
• acceptance-test-generator → work-planner information passing section in orchestration guide

Changed

Frontend Workflow Enhancements

• Orchestration flow step counts updated: large scale 11→13, medium scale 7→9 (frontend/fullstack)
• front-plan command restructured with test skeleton generation step
• implement command task execution section renamed and enhanced
• documentation-criteria skill updated with UI Spec in creation matrix, storage paths, diagram requirements, and templates
• skills-index.yaml updated with playwright/e2e tags and new references

Dependencies Modernization

• Remove 11 unused devDependencies: react, react-dom, @types/react, @types/react-dom, @vitejs/plugin-react, @testing-library/react, @testing-library/jest-dom, jsdom, c8, ts-node, ts-prune
• Remove 4 unused scripts: build:frontend, dev:frontend, preview, check:unused:all
• Migrate ts-prune → knip for unused export detection (check:unused now runs knip --include exports)
• Upgrade Biome v1 → v2 (schema migration: organizeImports → assist.actions.source, include/ignore → includes with ! prefix)
• Upgrade vitest v3 → v4, @vitest/coverage-v8 v3 → v4, @vitest/ui v3 → v4
• Upgrade vite v5 → v6, @types/node v20 → v22
• Add knip.json configuration

ESM Migration

• Migrate project from CommonJS to ES Modules ("type": "module" in package.json)
• Update tsconfig.json: target ES2020 → ES2022, module commonjs → Node16
• Convert all scripts/ and bin/ files from CJS (require/module.exports) to ESM (import/export)
• Add node: protocol to all Node.js builtin imports

Agent/Skill Environment Independence (en/ja)

• quality-fixer: replace ts-prune reference with generic "unused export detection tool"
• quality-fixer-frontend: replace hardcoded build:frontend with auto-detect pattern from package.json
• technical-spec skill: replace hardcoded script names with environment-independent descriptions
• frontend/technical-spec skill: replace hardcoded build commands with auto-detect pattern

Documentation

• Remove build:frontend and preview from README script lists (en/ja)

Removed

• .tsprunerc configuration file
• scripts/check-unused-exports.js (replaced by knip)

[1.16.3] - 2026-03-05

Fixed

• Replace deprecated TodoWrite with Tasks API (TaskCreate/TaskUpdate/TaskList) across all agent definitions, command files, and CLAUDE rule files (en/ja)
• Add missing TaskCreate/TaskUpdate to tools list for code-reviewer, design-sync, and integration-test-reviewer agents (en/ja)

[1.16.2] - 2026-03-03

Changed

Orchestrator Role Clarity (en/ja)

• Replace prohibition-style delegation ("NEVER investigate yourself") with positive role definition across build, design, implement, plan, update-doc commands: "your role is to invoke sub-agents, pass data between them, and report results"

reverse-engineer (en/ja) — Execution Accuracy

• Add Orchestrator Definition section with structured JSON passing and code-reading delegation
• Replace prose loop headers with explicit FOR ... (sequential, one unit at a time) notation (Steps 2-5, 7-10)
• Add prerequisite annotation to Step 3 (dependency on Step 2 output)
• Expand Step 5 revision prompt: replace vague "apply corrections" with severity-based handling (critical/important/recommended)
• Expand Step 10 from single-line reference to full backend + frontend revision prompts with severity-based handling
• Add 7a./7b. labels and sequential execution note for fullstack Design Doc generation
• Remove redundant "Context Passing" line (moved to Orchestrator Definition)

review (en/ja)

• Add $STEP_2_OUTPUT storage for code-reviewer results
• Replace vague re-validation ("measure improvement") with prior-issue-tracking re-validation using $STEP_2_OUTPUT

update-doc (en/ja)

• Add $STEP_5_OUTPUT storage for document-reviewer results
• Expand revision loop: replace single-line description with full sub-agent prompt including severity-based handling and $STEP_5_OUTPUT reference

Agents — Reverse Engineering Accuracy (en/ja)

• document-reviewer: add as-is implementation document review rule — verify code-observable behaviors are stated as facts, flag speculative language on deterministic behavior
• prd-creator: add Language Standard rule — code as SSoT, definitive form for observable behavior, "Undetermined Items" only for genuinely undeterminable claims
• prd-creator: add Inferred verification rule — attempt code reading before classifying claims as Inferred

[1.16.1] - 2026-02-23

Added

• /create-skill command (en/ja) for guided skill creation with interactive dialog and optimization
• skill-optimization skill (en/ja) with BP-001~008 content patterns and 9 editing principles, split into SKILL.md + references/ for context isolation
• skill-creator agent (en/ja) for optimized skill file generation
• skill-reviewer agent (en/ja) for skill quality evaluation with A/B/C grading

Changed

• /refine-skill command (en/ja): replace inline Three-Pass Review with skill-reviewer agent delegation

Fixed

• skills-index.yaml (en/ja): sync sections with actual SKILL.md H2 headings across coding-standards, technical-spec, project-context, documentation-criteria, subagents-orchestration-guide, frontend/technical-spec, frontend/typescript-rules (ja), frontend/typescript-testing (ja)

[1.16.0] - 2026-02-14

Added

Cross-Layer Orchestration

• Add cross-layer orchestration section to subagents-orchestration-guide SKILL.md (en/ja): design phase extensions, layer context prompts, vertical slice work planning, layer-aware agent routing table
• Add layer-aware task file naming convention to task-decomposer (en/ja): *-backend-task-* / *-frontend-task-* pattern for automatic agent routing
• Add Design Phase section to implement command (en/ja) for cross-layer scope detection
• Add layer-aware routing annotations to implement and build commands (en/ja)

Fullstack Reverse Engineering

• Add fullstack design option to reverse-engineer command (en/ja): per-unit backend + frontend Design Doc generation
• Add technicalProfile field to scope-discoverer output: primaryModules, publicInterfaces, dataFlowSummary, infrastructureDeps

Changed

scope-discoverer (en/ja) — Unified Single-Pass Discovery

• Remove scope_type parameter (previously prd | design-doc dual mode)
• Unify discovery sources into single table with Perspective column (user-value / technical / both)
• Add Granularity Criteria section with split/merge signals based on vertical slice principles
• Add implementation-approach to required skills
• Remove Prohibited Actions section (all items already covered by positive statements elsewhere)

reverse-engineer (en/ja) — Fullstack Integration

• Step 6: Change from scope-discoverer re-invocation to scope mapping (reuse Step 1 results)
• Step 7: Merge Standard/Fullstack mode split into unified per-unit flow
• Unify component terminology to unit throughout

Other Changes

• Support multiple Design Docs in work-planner required info and quality checklist (en/ja)
• Support multiple Design Docs in plan-template (en/ja)
• Update README (en/ja): remove /front-reverse-design, update mermaid diagram for scope reuse, update scope-discoverer description

Removed

• /front-reverse-design command (en/ja) — absorbed into /reverse-engineer with fullstack option

[1.15.1] - 2026-02-13

Changed

CLAUDE.md

• Restructure CLAUDE.md (en/ja) to linear step-based flow (Step 1-4), eliminating redundant sections
• Remove all inline "Reason:" annotations to reduce context token consumption
• Remove TypeScript-specific content (already covered by typescript-rules skill)
• Consolidate rule-advisor field references into a single location (Step 1)
• Add Quality Standard and Session Initialization sections
• Add quality-fixer and rule-advisor as explicit quality verification paths (Step 4)
• Convert auto-stop triggers to table format for scannability
• Specify concrete tool names (AskUserQuestion) for approval workflow
• Clarify ./tmp/ as project-root-relative directory

Commands

• Modernize refine-skill command (en/ja): Step-based structure, TodoWrite integration, completion criteria checklist, condensed 3-pass review (207→85 lines)
• Modernize sync-skills command (en/ja): TodoWrite integration, IF-THEN evaluation sequence, error handling table, Glob-based scanning instead of bash find (117→75 lines)
• Update task command (en/ja): add "Task Essence" as first metacognition item, rename rules→skills to align with CLAUDE.md
• Fix diagnose command (en/ja): migrate selectedRules field references to selectedSkills matching task-analyzer output format
• Modernize project-inject command (en/ja) with AskUserQuestion-based collection, dual output targets, and verification checklist

Skills

• Redesign project-context SKILL.md (en/ja) as AI-only prerequisite template, removing tech stack and implementation principles
• Sync active .claude/skills/project-context/SKILL.md with language-specific template

Documentation

• Update README (en/ja): /task description (rule→skill), project-inject description, skills table, FAQ, script reference terminology
• Rewrite quickstart guides (en/ja) to remove stale references (docs/rules/, /sync-rules, npx github:...)
• Update use-cases guides (en/ja): /task process steps and description, /sync-rules→/sync-skills, /refine-rule→/refine-skill, docs/rules→.claude/skills/ paths, rules→skills customization terminology
• Rename rule-editing-guide → skills-editing-guide (en/ja): full terminology migration from "rules" to "skills", add Skills Structure & Best Practices section with official docs links, Progressive Disclosure pattern, description/splitting guidelines, and common mistakes

[1.15.0] - 2026-02-12

Added

• npx create-ai-project update command to update agent definitions, commands, skills, and AI rules in existing projects
• .create-ai-project.json manifest file for version tracking
• Ignore mechanism (--ignore, --unignore) to protect user-customized files from being overwritten
• --dry-run flag to preview updates without applying changes
• This CHANGELOG file

[1.14.15] - 2026-02-10

Added

• Bash tool to 11 agents (both ja/en) that were missing shell command access, including investigator, verifier, requirement-analyzer, prd-creator, technical-designer, document-reviewer, code-reviewer, code-verifier, integration-test-reviewer, scope-discoverer, and technical-designer-frontend

[1.14.14] - 2026-02-04

Added

• Six design document quality improvements: Applicable Standards, Code Inspection Evidence, Data Representation Decision, and Field Propagation Map sections in design template
• Standards Identification Gate in technical designer workflow
• Gate-based review structure (Gate 0 / Gate 1) in document reviewer

[1.14.13] - 2026-02-02

Added

• /update-doc command for updating existing design documents (Design Doc / PRD / ADR) with built-in review workflow

Changed

• Unified command frontmatter format by removing redundant name field

[1.14.12] - 2026-01-23

Changed

• Unified Task tool invocation format across all command files
• Removed deprecated "Think deeply/harder" expressions
• Added Orchestrator Definition to add-integration-tests command

Fixed

• Heading hierarchy in review command

[1.14.11] - 2026-01-21

Changed

• Improved add-integration-tests command format with explicit Task tool invocation blocks, Role declaration, and expected output definitions

Fixed

• Incorrect subagents-orchestration-guide reference in Japanese diagnose command

[1.14.10] - 2026-01-20

Changed

• Converted requirement-analyzer output to mandatory JSON format with scopeDependencies and confidence fields
• Improved scale determination with Grep tool for accurate file count estimation
• Added step counts to flow headings and explicit AskUserQuestion requirements at Stop points

[1.14.9] - 2026-01-16

Fixed

• Stabilized agent JSON output by adding explicit "JSON format is mandatory" directive to code-verifier, scope-discoverer, investigator, and verifier agents

[1.14.8] - 2026-01-14

Added

• add-integration-tests command to generate integration/E2E test skeletons from Design Doc acceptance criteria
• front-reverse-design command for reverse engineering PRD and Design Doc from frontend codebase
• front-review command for Design Doc compliance validation

Changed

• Simplified task-decomposer agent and enhanced subagents-orchestration-guide

Fixed

• Review command inconsistencies between en/ja versions

[1.14.7] - 2026-01-12

Changed

• Added Orchestrator Definition with explicit work-to-agent mapping in Execution Method sections
• Added Workflow Overview diagrams and Scope Boundaries to design commands
• Replaced "Avoid" sections with Responsibility Boundary pattern to prevent pink elephant problem

[1.14.6] - 2026-01-09

Added

• Complexity gate for over-engineering detection in design review process with complexity_level and complexity_rationale fields

[1.14.5] - 2026-01-08

Changed

• Enhanced diagnostic framework with multi-cause support, confidence loop (up to 2 iterations), rule-advisor integration, investigation focus points, design gap escalation, and residual risk documentation

[1.14.4] - 2026-01-06

Changed

• Simplified orchestrator role definitions using positive-form instructions
• Streamlined diagnose flow to always execute investigator -> verifier -> solver

Removed

• "Pink elephant" negative constraints from orchestrator commands

[1.14.3] - 2026-01-05

Added

• WebSearch tool to solver agent for verifying generated solutions against project rules and best practices

[1.14.2] - 2025-12-31

Fixed

• Skill descriptions not displaying correctly in Claude Code due to YAML folded style (>) being parsed incorrectly
• Converted 26 skill files to single-line description format

[1.14.1] - 2025-12-30

Changed

• Enhanced skill descriptions with action verbs and "Use when:" triggers for better LLM discoverability
• Enhanced agent descriptions with trigger words and PROACTIVELY markers for auto-activation
• Updated 26 skill files and 38 agent files

[1.14.0] - 2025-12-26

Added

• /reverse-engineer command to generate PRD and Design Doc from existing codebases
• scope-discoverer agent for documentation scope discovery using multi-source triangulation
• code-verifier agent for document-code consistency validation

Changed

• Updated document-reviewer with JSON output format and prior context handling
• Updated prd-creator with external scope support and confidence gating

[1.13.1] - 2025-12-21

Changed

• Enhanced investigator with problem type classification, causal chain tracking, causeCategory, comparisonAnalysis, and impactAnalysis
• Enhanced verifier with counter-evidence weighting and scopeValidation
• Added Step 0 (problem structuring) to diagnose command

[1.13.0] - 2025-12-21

Added

• /diagnose command for root cause analysis with 5-step flow
• investigator agent for comprehensive problem investigation with evidence matrix
• verifier agent for investigation verification using ACH and Devil's Advocate methods
• solver agent for solution derivation with tradeoff analysis

[1.12.1] - 2025-12-19

Fixed

• Excluded generated Claude Code files (.claude/agents/, .claude/commands/, .claude/skills/) from npm package

[1.12.0] - 2025-12-19

Added

• Skills system: all rules migrated to .claude/skills-{en,ja}/ as SKILL.md files
• task-analyzer skill for metacognitive task analysis
• subagents-orchestration-guide skill for agent coordination

Changed

• All 16 agents updated with skills: frontmatter
• rule-advisor JSON output field renamed: selectedRules to selectedSkills with ~80% size reduction

Removed

• docs/rules-en/ and docs/rules-ja/ directories (migrated to skills)
• docs/guides/ directory (moved to skills)
• Template files from docs/ (moved to .claude/skills/*/references/)

[1.11.3] - 2025-12-18

Added

• Positive criteria to suppress test antipatterns: "Valid test: Expected value != Mock return value" criterion for detecting tautology tests

[1.11.2] - 2025-12-11

Changed

• Renamed package to create-ai-project (previously ai-coding-project-boilerplate)
• Now supports npx create-ai-project and npm init ai-project shorthands

[1.11.1] - 2025-12-07

Added

• "Decision Details" section (Why now / Why this / Known unknowns / Kill criteria) in ADR template
• "Design Summary (Meta)" block with risk level, constraints, and unknowns in Design Doc template
• "Failure scenario review" perspective in document-reviewer agent

[1.11.0] - 2025-12-05

Added

• integration-test-reviewer agent for verifying test implementation quality
• design-sync agent for ensuring Design Document consistency
• EARS notation for acceptance criteria (Event-Action-Response-State format)
• Property-based testing support with fast-check integration
• check:code script consolidating code quality checks

Changed

• Clarified quality-fixer phase structure by removing ambiguous Phase 6

[1.10.0] - 2025-12-03

Added

• Package manager flexibility: support for npm, bun, and pnpm via packageManager field in package.json
• Claude Code automatically detects and uses the configured package manager

[1.9.3] - 2025-11-30

Removed

• Ambiguous "Proactive Use of Specialized Agents" section from CLAUDE.md to reduce token count and improve clarity

[1.9.2] - 2025-11-30

Changed

• Enhanced TodoWrite integration with rule-advisor output (firstActionGuidance, taskEssence, warningPatterns)
• Replaced prohibition-focused expressions with positive action alternatives (Pink Elephant Problem fix)

[1.9.1] - 2025-11-24

Changed

• Simplified Fail-Fast principle documentation from 16 lines to 4 lines in coding-standards.md
• Moved implementation specifics to language-specific rules (typescript.md, frontend/typescript.md)

[1.9.0] - 2025-11-07

Added

• ROI-based test generation with 3-layer quality filtering (60-75% test reduction)
• Universal coding-standards.md consolidating duplicated Frontend/Backend rules (~760 lines removed)
• Dynamic year detection for future-proof latest information searches

Changed

• Agent prompt optimization reducing context pollution (~130 lines net reduction)
• Simplified AC scoping in technical-designer

Fixed

• Initialization script template file detection logic

[1.8.0] - 2025-10-26

Added

• Frontend development infrastructure with React support
• 3 specialized frontend agents: technical-designer-frontend, task-executor-frontend, quality-fixer-frontend
• 3 frontend slash commands: /front-design, /front-plan, /front-build
• Frontend-specific rules for props-driven component design, React Testing Library + MSW patterns

Changed

• Technology-agnostic approach: removed version-specific dependencies (React 19, Vite, Lighthouse)

[1.7.17] - 2025-10-18

Fixed

• Clarified rule-advisor agent must be invoked using the Task tool (was incorrectly attempted via Python or other tools)

[1.7.16] - 2025-10-13

Added

• 3-stage impact analysis process (Discovery, Understanding, Identification) for implementation completeness
• Unused code deletion rule requiring immediate decision (implement or delete, no deferral)
• Existing code deletion decision flow with clear criteria

[1.7.15] - 2025-10-12

Changed

• Optimized all rule documentation to save ~2,300 tokens of context consumption
• Removed meta descriptions, simplified YAML structures, condensed code comments
• Conservative approach: only changes with 0% execution accuracy impact applied

[1.7.14] - 2025-10-10

Changed

• Improved Acceptance Criteria scoping guidelines optimized for autonomous LLM implementation in CI/CD
• Simplified "Out of Scope" into 3 categories: External Dependencies, Non-Deterministic in CI, Implementation Details
• Added "AC Scoping for Autonomous Implementation" section to technical-designer

[1.7.13] - 2025-10-07

Removed

• Unused Vertical Slice and Hybrid Progressive architecture rules to prevent context pollution and AI precision loss (-1,306 lines, -7 files)

[1.7.12] - 2025-09-30

Changed

• Optimized orchestration workflow for Claude Sonnet 4.5 cognitive patterns
• Added explicit Main AI verification step between acceptance-test-generator and work-planner to prevent step skipping
• Introduced "Information Bridge Confirmation" approach for automated file verification

[1.7.11] - 2025-09-28

Fixed

• Excessive test generation caused by ambiguous rule descriptions (~97% reduction in false positives)
• Removed ambiguous terms like "operational continuity necessity" that caused LLM misinterpretation

[1.7.10] - 2025-09-24

Added

• Integration point analysis for cross-feature E2E testing with impact levels (High/Medium/Low)
• Fail-fast principles for better error detection, replacing unconditional fallback implementations

[1.7.9] - 2025-09-09

Changed

• Renamed e2e-test-generator to acceptance-test-generator for broader coverage
• E2E tests now run only when implementation is complete, preventing Red-phase blocking
• Clear separation between unit tests (Phase 0), integration tests, and E2E tests (final validation)

[1.7.8] - 2025-09-08

Added

• /refine-rule command for optimizing project rules with three-pass review process
• Quick Start Guide for 5-minute onboarding
• Use Cases Quick Reference for daily workflow patterns

[1.7.7] - 2025-09-05

Changed

• Improved /implement command with phase-based execution flow for better stability
• Added crash prevention measures: rule-advisor recursion prevention, high-risk agent protection

[1.7.6] - 2025-08-31

Added

• /sync-rules command for automatic rule metadata synchronization after edits

Changed

• Enhanced agent implementation consistency with Design Doc compliance
• Improved TypeScript test design standards and type safety

Fixed

• Metadata inconsistencies in typescript-testing rules

[1.7.5] - 2025-08-30

Added

• Enhanced unused exports detection script (check-unused-exports.js) with auto-removal capability
• npm run check:unused now uses improved script; npm run check:unused:all for raw ts-prune output

[1.7.4] - 2025-08-29

Fixed

• Subagents now operate independently without referencing parent CLAUDE.md configuration
• Task executor checkbox update functionality restored with precise pattern matching

[1.7.3] - 2025-08-28

Fixed

• task-executor checkbox update functionality broken in v1.7.0
• Restored concrete 3-step progress update instructions with [MANDATORY] enforcement keywords

[1.7.2] - 2025-08-28

Added

• /project-inject slash command for interactive project context configuration
• Guided setup process replacing manual placeholder editing in project-context.md

[1.7.1] - 2025-08-26

Fixed

• Incorrect section references in quality-fixer agent definitions (both ja/en)

[1.7.0] - 2025-08-24

Added

• e2e-test-generator agent for automated integration test skeleton creation
• Structured task decomposition with granular progress tracking in task-executor
• E2E verification procedures in Design Document and Work Plan templates

Changed

• Enhanced quality-fixer with comprehensive error handling and pre-commit hook retry mechanisms
• Improved technical-designer ADR decision-making process and Design Document structure
• Enhanced work-planner with Design Document alignment and phase-based implementation

[1.6.7] - 2025-08-20

Added

• Pattern 5 anti-pattern "Insufficient Existing Code Investigation" to prevent duplicate implementations
• Pre-implementation existing code investigation checks for all AI agents

[1.6.6] - 2025-08-19

Changed

• Clarified orchestrator commit execution responsibilities in sub-agents documentation
• Explicit mention of using Bash tool for git commit execution in autonomous mode

[1.6.5] - 2025-08-17

Added

• Modern TypeScript best practices: branded types, functional programming (Effect-TS/fp-ts), modern DI patterns, satisfies operator, template literal types
• Comprehensive Red-Green-Refactor TDD methodology documentation

Changed

• Fixed systematic inconsistencies between Japanese and English documentation
• Restructured Design Document template (269 lines, 48 sections)

[1.6.4] - 2025-08-15

Changed

• Simplified document-reviewer by removing unused iteration-based approach and non-deterministic review patterns

Removed

• References to non-existent document-fixer sub-agent

[1.6.3] - 2025-08-15

Fixed

• Missing implementation-approach.md references in technical-designer and work-planner agents
• Progress checkbox update omissions in task-executor with concrete Edit tool instructions

[1.6.2] - 2025-08-10

Changed

• Enhanced English agent definitions with improved capabilities and constraints
• Standardized template documentation across ADR, Design Documents, Plans, and PRD
• Streamlined architecture process by removing obsolete documentation

Added

• New architecture implementation approach guidelines

[1.6.1] - 2025-08-09

Added

• L1/L2/L3 verifiability levels for systematic task validation
• Vertical/horizontal slice strategy for optimal task ordering
• Composite review mode for multi-angle document verification in single execution

Removed

• Suspended document-fixer agent

[1.6.0] - 2025-08-08

Added

• code-reviewer agent for comprehensive automated code review (security, performance, maintainability)
• /review command for triggering code reviews

Changed

• Optimized agent definitions for improved execution accuracy
• Enhanced error handling in quality-fixer and task execution flow in task-executor

[1.5.1] - 2025-08-07

Fixed

• Stability issues in build command's autonomous execution mode
• Structured response processing for reliable readyForQualityCheck: true signal detection
• Quality gate reliability ensuring all checks execute before task completion

[1.5.0] - 2025-08-07

Added

• Technical information verification in document-reviewer with WebSearch integration
• Reverse engineering mode in prd-creator for existing implementations
• Latest technology research in technical-designer with automatic source citation

Changed

• Improved PRD determination logic with three modes: create, update, and reverse-engineer
• Fixed scale assessment terminology inconsistencies across agents

[1.4.4] - 2025-08-05

Changed

• Removed rule-advisor dependencies from subagent definitions for improved maintainability
• Simplified task-executor configuration by removing ambiguous mandatory check items

[1.4.3] - 2025-08-04

Changed

• Adapted to Claude Code behavior change where subagents can no longer invoke other subagents through Task tool
• Removed Task tool from all 18 subagent definitions and replaced with direct rule file references

Deprecated

• document-fixer marked as suspended due to Task tool limitations

[1.4.2] - 2025-08-03

Added

• NPX support for quick project setup: npx github:shinpr/ai-coding-project-boilerplate my-project
• Interactive configuration with language selection and automatic dependency installation

[1.4.1] - 2025-08-03

Changed

• Restructured reference management: removed redundant references from rule files, centralized in rules-index.yaml
• Enhanced rule-advisor from "minimal necessary" to "comprehensive quality" approach
• Achieved ~18% reduction in rule file sizes for better context window utilization

[1.4.0] - 2025-08-03

Added

• Dynamic rule selection system with rules-index.yaml providing metadata for each rule file
• rule-advisor subagent for intelligent rule selection based on task requirements
• Mandatory integration between TodoWrite and rule-advisor

Changed

• Rule content optimized: ~15-18% reduction across all languages
• Enhanced agent behavior controls with automatic stops for large-scale changes

Removed

• canonical-phrases.md (content integrated into other rule files)

[1.3.1] - 2025-08-02

Added

• Self-reflection questions in pre-implementation planning (task essence, rule-based prioritization, failure pattern recognition)
• Task command files for both Japanese and English environments

[1.3.0] - 2025-08-01

Added

• Multi-language support with dynamic switching between Japanese and English
• scripts/set-language.js for language management with .claudelang config file
• Complete bilingual file structure for all documentation, agents, commands, and templates
• npm scripts: lang:ja, lang:en, lang:status

[1.2.4] - 2025-08-01

Changed

• Improved task-executor execution precision with clear implementation permissions and action principles
• Fixed /plan command to focus on work plan creation only (no longer auto-starts implementation)
• Enhanced /build command to manage full cycle from task decomposition to quality assurance

[1.2.3] - 2025-08-01

Added

• 5 custom slash commands: /onboard, /design, /plan, /implement, /build
• Each command covers a distinct development phase with clear boundaries

[1.2.2] - 2025-07-31

Fixed

• task-executor autonomy: no longer repeats unnecessary confirmation prompts
• Duplicate task execution prevention for already-completed tasks
• Clarified task-executor instruction method (task file path only)

[1.2.1] - 2025-07-31

Changed

• Introduced 2-stage document fixing process (content improvement then AI optimization)
• Moved consistency verification to final phase for better new-inconsistency detection
• Added concrete criteria for ambiguous expression replacement and term unification

[1.2.0] - 2025-07-30

Added

• document-fixer subagent for automated document correction based on review results
• Parallel execution of document-reviewer from multiple review perspectives

Changed

• Enhanced document-reviewer with more structured review process

[1.1.2] - 2025-07-30

Fixed

• Recursive call issue in CLAUDE.md where subagents reading the file triggered infinite recursion
• Removed all subagent-related descriptions from CLAUDE.md to resolve circular dependency

[1.1.1] - 2025-07-29

Changed

• Optimized CLAUDE.md: added project-specific rule priority, changed "guidelines" to "rules" for stronger enforcement
• Removed ~75 lines duplicated between CLAUDE.md and rule files
• Introduced @ notation for unified subagent invocation
• Removed proactive execution conditions from quality-fixer to eliminate improper inter-agent dependencies

[1.1.0] - 2025-07-28

Added

• Requirements change detection in sub-agents guide
• Autonomous execution phase clarification (quality assurance / error fix / implementation)

Changed

• Comprehensive improvements to all 8 subagents (document-reviewer, prd-creator, quality-fixer, requirement-analyzer, task-decomposer, task-executor, technical-designer, work-planner)
• Simplified TodoWrite tool usage in ai-development-guide.md

[1.0.0] - 2025-07-28

Added

• 8 specialized subagents (prd-creator, technical-designer, work-planner, requirement-analyzer, task-decomposer, task-executor, quality-fixer, document-reviewer)
• Interactive mode for prd-creator and requirement-analyzer
• Mermaid diagram guidelines for prd-creator, technical-designer, and work-planner

Changed

• Sub-agents guide fully revised as a practical LLM guide

[0.1.0] - 2025-07-27

Added

• Initial release: Claude Code optimized TypeScript project boilerplate
• 6 development rule files covering technical specs, TypeScript, testing, and AI development
• Layered architecture patterns (Vertical Slice, Layered, Hybrid Progressive)
• Quality assurance system with 6-phase checks
• Biome linter/formatter, Vitest test framework, Husky + lint-staged
• ADR process, work plan templates, and sub-agent utilization guide