All notable changes to this project will be documented in this file.
Release Notes:
- Add Greek (el) language support, bringing total to 13 languages (#121)
- Update GitHub Copilot prompt files to replace deprecated
modeattribute withagent(#118) - Fix registry.ts with review improvements (#107)
- Add AI-Assisted SDD book reference to documentation (#109)
- Refine recommended OpenAI models for Codex CLI, Cursor, GitHub Copilot, and Windsurf agents to prioritize
gpt-5.1-codex medium/high, keepinggpt-5.1 medium/highas a general-purpose fallback.
-
Align DEV_GUIDELINES-related tests with the stricter language-handling rules introduced in v2.0.2 so
npm testpasses cleanly for v2.0.3. -
PRs: #104
- Align templates, rules, and prompts with GPT-5.1 by updating recommended OpenAI model names for Codex CLI, Cursor, GitHub Copilot, and Windsurf agents to
GPT-5.1 high or medium. - Tighten language handling so all generated Markdown (requirements, design, tasks, research, validation) uses the spec’s target language (
spec.json.language) and defaults to English (en) when unspecified. - Make EARS patterns and requirements traceability more consistent by keeping EARS trigger phrases (
When,If,While,Where,The system shall,The [system] shall) as fixed English fragments, localizing only the variable slots, and enforcing numeric requirement IDs across all phases (e.g.Requirement 1,1.1,2.3) with fast failure when IDs are missing or invalid instead of falling back to free-form labels. - PRs: #102
- Consolidates every feature shipped in 2.0.0-alpha.1〜alpha.6 and promotes them to
npx cc-sdd@latest. - Adds validation commands, Research.md, steering/memory upgrades, and 7-agent / 13-language parity.
- For migration steps, see
docs/guides/migration-guide.md(referenced from release notes as well).
- Parallel task analysis by default in spec-tasks command (#89)
- Automatic
(P)marker for parallel-executable tasks - New
--sequentialflag to opt-out of parallel analysis - New rule file:
tasks-parallel-analysis.mdfor identifying parallel tasks
- Automatic
- Research.md template for spec-driven workflow
- Separates discovery findings and architectural investigations from
design.md - Captures research logs, architecture pattern evaluations, and design decisions
- Provides structured format for documenting trade-offs and rationale
- Separates discovery findings and architectural investigations from
- Guidelines for excluding agent tooling directories from steering docs
- Prevents
.claude/,.cursor/,.codex/etc. from being analyzed
- Prevents
- Claude Code Subagents mode for context optimization (#74)
- Delegate SDD commands to dedicated subagents to preserve main conversation context
- Improve session lifespan by isolating command-specific context
- Specialized system prompts for each command type
- 12 commands + 9 Subagent definitions
- Windsurf IDE support with complete workflow integration
.windsurf/workflows/directory with 11 workflow files- AGENTS.md configuration for optimization
--windsurfCLI flag
- Codex CLI official support with 11 prompts in
.codex/prompts/ - GitHub Copilot official support with 11 prompts in
.github/prompts/
/kiro:validate-gap- Analyze implementation gap between requirements and existing codebase/kiro:validate-design- Validate design compatibility with existing architecture/kiro:validate-impl- Validate implementation against requirements, design, and tasks
- Interactive CLI installer with guided setup (#70)
- Organized file display by Commands / Project Memory / Settings categories
- Interactive project memory handling (overwrite/append/keep)
- Comprehensive documentation
- npm badges for version tracking (#86)
- Unified template structure - removed
os-mac/os-windowsdirectories in favor of singlecommands/structure - All templates now use actual extensions (
.md,.prompt.md,.toml) - Steering now functions as project-wide rules/patterns/guidelines (Project Memory)
- Enhanced steering system loading all documents under
steering/directory
- Enhanced steering system loading all documents under
- Shared settings bundle in
{{KIRO_DIR}}/settingsfor cross-platform customization
- Redesigned all 11 Spec-Driven commands (
spec-*,validate-*,steering*) with improved context - Enhanced task generation guidelines with parallel execution criteria
- Improved design template with discovery process guidelines
- Updated spec-design workflow to leverage new research.md template
- Streamlined tasks.md template structure
- Updated EARS format to use lowercase syntax (#88)
- Changed from "WHILE/WHEN/WHERE/IF" to "while/when/where/if"
- Improved readability and consistency
- Clarified template customization instructions (#85)
- Updated installation documentation for better clarity (#87)
- Reorganized documentation structure
- Renamed
docs/CHANGELOG/todocs/RELEASE_NOTES/ - Separated technical changelog from marketing-focused release notes
- Added cross-references between CHANGELOG and Release Notes
- Renamed
- Automated GitHub issue lifecycle management (#80)
- Auto-close stale issues after 10 days of inactivity
- Configurable stale detection workflow
- English-only workflow messaging (#81)
- Centralized agent metadata into registry (#72)
- Template structure standardization across all agents
- Manifest definitions for new directory layouts
- Template parameter replacement across platforms
- OS-specific command handling for Windows environments
- OS-specific template directories (
os-mac,os-windows) - Deprecated Claude documentation files
- Duplicate CLAUDE.md files
- Unused documentation artifacts
- Template Structure: OS-specific directories removed. Use unified templates in
.kiro/settings/templates/ - Steering: Now loads entire
steering/directory instead of single file - File Extensions: Templates use actual extensions (
.md,.prompt.md,.toml) - Command Count: Expanded from 8 to 11 commands (3 validation commands added)
See the comprehensive Migration Guide for detailed upgrade instructions, including:
- Step-by-step migration procedures
- Breaking changes explained
- Template and steering migration
- Troubleshooting common issues
For release storytelling, refer to
docs/RELEASE_NOTES/*. This changelog keeps the technical diff only.
- Parallel task analysis features (included in v2.0.0)
- Research.md template (included in v2.0.0)
- npm
nextversion badge in README files (#86)
- Updated EARS format to use lowercase syntax (#88)
- Changed from "WHILE/WHEN/WHERE/IF" to "while/when/where/if"
- Improved readability and consistency
- Updated installation documentation for better clarity (#87)
Related PRs:
- #88 - Update EARS format to lowercase syntax
- #87 - Clarify installation
- #86 - Add npm next badge to README files
- Comprehensive customization guide with 7 practical examples (#83)
- Template customization patterns
- Agent-specific workflow examples
- Project-specific rule examples
- Complete command reference documentation (#83)
- Detailed usage for all 11
/kiro:*commands - Parameter descriptions and examples
- Detailed usage for all 11
Related PRs:
- #83 - Add customization guide and command reference
- #84 - Customization guide review suggestions
- #85 - Clarify template customization instructions
- Automated GitHub issue lifecycle management (#80)
- Auto-close stale issues after 10 days of inactivity
- Configurable stale detection workflow
- English-only workflow messaging (#81)
- Updated stale detection period to 10 days
- Improved GitHub Actions workflow for issue management
Related PRs:
- Windsurf IDE agent definition, manifest, and workflow templates so
npx cc-sdd@next --windsurfinstalls.windsurf/workflows/and AGENTS.md alongside shared settings. realManifestWindsurfvitest coverage that exercises dry-run and apply flows across macOS/Linux runtimes.--windsurfCLI alias support and accompanying argument parser tests.
- Updated completion guides and recommended model messaging to include Windsurf-specific guidance.
- Refreshed root README,
tools/cc-sdd/README*, anddocs/README/README_{en,ja,zh-TW}.mdwith Windsurf setup instructions and manual QA steps.
- Claude Code Subagents mode for context optimization (#74)
- Delegate SDD commands to dedicated subagents to preserve main conversation context
- Improve session lifespan by isolating command-specific context
- Specialized system prompts for each command type
- Related issue: #68
- CHANGELOG.md at root following Keep a Changelog format
- Release Notes documentation structure in
docs/RELEASE_NOTES/- Japanese version (RELEASE_NOTES_ja.md)
- English version (RELEASE_NOTES_en.md)
- Reorganized documentation structure
- Renamed
docs/CHANGELOG/todocs/RELEASE_NOTES/ - Separated technical changelog from marketing-focused release notes
- Added cross-references between CHANGELOG and Release Notes
- Renamed
- Improved Claude Code agent templates with updated recommendations
- Centralized agent metadata into registry (#72)
- Deprecated Claude documentation files
- Duplicate CLAUDE.md files
- Unused documentation artifacts
Related PRs:
- #74 - Add Claude Code Subagents mode
- #73 - Add CLAUDE.md documentation
- #72 - Refactor agent metadata into central registry
- Interactive CLI installer with guided setup (
npx cc-sdd@latest)- Organized file display by Commands / Project Memory / Settings categories
- Interactive project memory handling (overwrite/append/keep)
- Codex CLI official support with 11 prompts in
.codex/prompts/ - GitHub Copilot official support with 11 prompts in
.github/prompts/ - Shared settings bundle in
{{KIRO_DIR}}/settingsfor cross-platform customization - Enhanced steering system loading all documents under
steering/directory
- Redesigned all 11 Spec-Driven commands (
spec-*,validate-*,steering*) with improved context - Unified template structure - removed
os-mac/os-windowsdirectories in favor of singlecommands/structure - All templates now use actual extensions (
.md,.prompt.md,.toml) - Steering now functions as project-wide rules/patterns/guidelines (Project Memory)
- Updated manifests and CLI with
--codex,--github-copilotflags
- Template structure standardization across all agents
- Manifest definitions for new directory layouts
- OS-specific template directories (
os-mac,os-windows)
Metrics:
- Supported Platforms: 6 (Claude Code, Cursor IDE, Gemini CLI, Codex CLI, GitHub Copilot, Qwen Code)
- Commands: 11 (6 spec + 3 validate + 2 steering)
Related PRs:
- Qwen Code AI assistant support (#64)
- Reuse gemini-cli templates to minimize code duplication
- Command directory:
.qwen/commands/kiro - QWEN.md template for project memory
- Bash command errors in steering templates (#62)
- Reverted to original bash one-liner style
- Maintained Windows compatibility
- Improved steering command templates (#60)
- Simplified custom files check logic using
ls + wc - Added
AGENTS.mdto project analysis section
- Simplified custom files check logic using
- Kiro IDE integration descriptions in READMEs (#61)
- Clarified spec portability to Kiro IDE
- Removed confusing command references
- Multi-language support for project memory documents (#59)
- Centralized development guideline strings by language
- Single templates with
DEV_GUIDELINESplaceholder
- Consolidated agent documentation templates
- Updated manifests and tests for new template structure
- Updated repository URL throughout the project
- Improved test coverage and fixed edge cases (#57)
- CLI messages and linux template mapping expectations
- Generated artifacts now properly ignored (.claude/, CLAUDE.md)
- Validation commands for brownfield development (#56)
/kiro:validate-gap- Analyze implementation gap between requirements and existing codebase/kiro:validate-design- Validate design compatibility with existing architecture/kiro:validate-impl- Validate implementation against requirements, design, and tasks
- Cursor IDE official support with 11 commands
- AGENTS.md configuration file for Cursor IDE optimization
- Windows template support for Gemini CLI with proper bash -c wrapping (#56)
- Command structure expanded from 8 to 11 commands
- Enhanced spec-design with flexible system flows and requirements traceability (#55)
- Improved EARS requirements template with better subject guidance
- Updated documentation for brownfield vs greenfield workflows
- Template parameter replacement across platforms
- OS-specific command handling for Windows environments
Metrics:
- Supported Platforms: 5 (Claude Code, Cursor IDE, Gemini CLI, Codex CLI, GitHub Copilot)
- Commands: 11 (6 spec + 3 validate + 2 steering)
- Documentation Languages: 3 (English, Japanese, Traditional Chinese)
Related PRs:
- #56 - Reorganize templates by OS and add Gemini CLI support
- #55 - Enhance technical design document generation
- #54 - Improve slash commands with individual arguments
- #52 - Add Cursor agent manifest and CLI support
- Multi-platform support for Spec-Driven Development
- Claude Code (original platform)
- Cursor IDE integration
- Gemini CLI with TOML configuration
- Codex CLI with GPT-5 optimized prompts
- cc-sdd npm package for easy distribution (#39)
- Complete CLI tool with
npx cc-sdd@latestinstallation - Template system supporting multiple platforms and OS variants
- 8 core commands for spec-driven workflow
- spec-init, spec-requirements, spec-design, spec-tasks
- spec-impl, spec-status
- steering, steering-custom
- Complete workflow redesign for spec-driven development
- Unified output format across all platforms
- Individual argument handling (
$1,$2) instead of$ARGUMENTS(#54)
- Context creation optimization in template processing (#45)
- Eliminated redundant
contextFromResolved()calls - Improved performance by 20-50% for template-heavy operations
- Eliminated redundant
Metrics:
- Supported Platforms: 4 (Claude Code, Cursor, Gemini CLI, Codex CLI)
- Commands: 8
- Documentation Languages: 3
Related PRs:
- #54 - Improve slash commands with individual arguments
- #52 - Add Cursor agent support
- #51 - Major enhancement of kiro commands
- #45 - Optimize context creation performance
- #43 - Add CI/CD workflow
- #42 - Refactor README structure
- #39 - Add gemini-cli integration and cc-sdd tool
- #37 - Release v1.0.0-beta.1
- #36 - Initial CLI tool release
-yflag for streamlined workflow approval- Skip requirement approval:
/kiro:spec-design feature-name -y - Skip requirement + design approval:
/kiro:spec-tasks feature-name -y
- Skip requirement approval:
- Argument hints in command input (
<feature-name> [-y]) - Custom Steering support in all spec commands
- Optimized command file sizes by 30-36%
- spec-init.md: 162→104 lines (36% reduction)
- spec-requirements.md: 177→124 lines (30% reduction)
- spec-tasks.md: 295→198 lines (33% reduction)
- Task structure optimization
- Section-based functional grouping
- Task granularity limits (3-5 sub-items, 1-2 hour completion)
- Unified requirements reference format
- Redundant explanations and template sections
- "Phase X:" prefixes in task organization
- Optimized CLAUDE.md file size from 150 to 66 lines
- Removed duplicate sections and verbose explanations
- Applied optimization across all language versions (Japanese, English, Traditional Chinese)
- "think" keyword to spec-requirements.md for better AI reasoning
- Interactive approval system for workflow phases
/kiro:spec-design: Prompts for requirements review confirmation/kiro:spec-tasks: Prompts for requirements + design review confirmation- Automatic spec.json updates on 'y' approval
- Enhanced specification generation quality
- Improved EARS format consistency in requirements.md
- Research & analysis process in design phase
- Requirements mapping and traceability in design.md
- TDD-optimized task structure in tasks.md
- Directory handling when
.kiro/steering/doesn't exist - Error messages improved for better clarity
- Simplified system design by removing redundant
progressfield - Reverted to original Kiro design philosophy for requirements generation
- Removed excessive "CRITICAL" and "MUST" language
- Focus on core functionality with iterative improvement
- Security guidelines and content quality guidelines
- Inclusion modes improvements (Always/Conditional/Manual)
- Detailed usage recommendations and guidance
- Enhanced
/kiro:steeringcommand to properly handle existing files - Improved steering document management
- Claude Code pipe bugs for more reliable execution
- Non-git environment compatibility
- Kiro IDE-style Spec-Driven Development system
- 3-phase approval workflow (Requirements → Design → Tasks → Implementation)
- EARS format requirement definition support
- Hierarchical requirement structure
- Automatic progress tracking and hooks
- Basic Slash Commands set
- Manual approval gates for quality assurance
- Specification compliance checking
- Context preservation functionality
- Initial project structure
- Repository: gotalab/cc-sdd
- npm Package: cc-sdd
- Release Notes:
- Documentation:
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.