A CLI toolkit for AI-assisted software development with structured phase templates and environment setup for Cursor and Claude Code.
- π― Phase-based Development: Structured templates for each stage of the software development lifecycle
- π€ AI Environment Setup: Automatic configuration for Cursor and Claude Code
- π Customizable Templates: Markdown-based templates with YAML frontmatter
- π Interactive CLI: User-friendly prompts with flag override support
- βοΈ State Management: Tracks initialized phases and configuration
# Using npx (no installation needed)
npx ai-devkit init
# Or install globally
npm install -g ai-devkit
Initialize AI DevKit in your project:
# Interactive mode (recommended)
ai-devkit init
# With flags
ai-devkit init --environment cursor --all
# Initialize specific phases
ai-devkit init --phases requirements,design,planning
This will:
- Create a
.ai-devkit.json
configuration file - Set up your AI development environment (Cursor/Claude Code)
- Generate phase templates in
docs/ai/
- Requirements: Problem understanding, requirements gathering, and success criteria
- Design: System architecture, data models, and technical design (include mermaid diagrams for architecture/data flow)
- Planning: Task breakdown, milestones, and project timeline
- Implementation: Technical implementation notes and code guidelines
- Testing: Testing strategy, test cases, and quality assurance
- Deployment: Deployment process, infrastructure, and release procedures
- Monitoring: Monitoring strategy, metrics, alerts, and observability
Initialize AI DevKit in your project.
Options:
-e, --environment <env>
: Specify environment (cursor|claude|both)-a, --all
: Initialize all phases at once-p, --phases <phases>
: Comma-separated list of specific phases
Examples:
# Interactive mode
ai-devkit init
# Initialize for Cursor with all phases
ai-devkit init --environment cursor --all
# Initialize specific phases
ai-devkit init --phases requirements,design,implementation
Add or update a specific phase template.
Examples:
# Interactive selection
ai-devkit phase
# Add specific phase
ai-devkit phase requirements
ai-devkit phase testing
After initialization, your project will have:
your-project/
βββ .ai-devkit.json # Configuration and state
βββ docs/
β βββ ai/
β βββ requirements/
β β βββ README.md
β βββ design/
β β βββ README.md
β βββ planning/
β β βββ README.md
β βββ implementation/
β β βββ README.md
β βββ testing/
β β βββ README.md
β βββ deployment/
β β βββ README.md
β βββ monitoring/
β βββ README.md
βββ [Environment-specific files]
βββ .cursor/
βββ rules/ # Project-specific rules (Markdown files)
β βββ ai-devkit.md
βββ commands/ # Custom slash commands (Markdown files)
βββ new-requirement.md
βββ code-review.md
βββ execute-plan.md
βββ writing-test.md
βββ update-planning.md
βββ check-implementation.md
βββ review-design.md
βββ review-requirements.md
βββ .claude/
βββ CLAUDE.md # Workspace configuration
βββ commands/ # Custom commands (Markdown files)
βββ new-requirement.md
βββ code-review.md
βββ execute-plan.md
βββ writing-test.md
βββ update-planning.md
βββ check-implementation.md
βββ review-design.md
βββ review-requirements.md
All templates are plain Markdown files with YAML frontmatter. You can customize them to fit your project's needs:
---
phase: requirements
title: Requirements & Problem Understanding
description: Clarify the problem space, gather requirements, and define success criteria
---
# Your custom content here
Templates are designed to provide structure while remaining concise and AI-friendly.
Generated files:
.cursor/rules/
: Project-specific rules as Markdown files (per Cursor documentation).cursor/commands/
: Custom slash commands as Markdown files (per Cursor documentation)
Available slash commands:
/new-requirement
: Complete workflow for adding a new feature from requirements to PR/code-review
: Structured local code review against design docs before pushing changes/execute-plan
: Walk a feature plan task-by-task with interactive prompts/writing-test
: Write unit/integration tests targeting 100% coverage/update-planning
: Update planning and task breakdown/check-implementation
: Compare implementation with design/review-design
: Review system design and architecture/review-requirements
: Review and summarize requirements
Each command is stored as a plain Markdown file in .cursor/commands/
and will automatically appear when you type /
in Cursor's chat input.
Generated files:
.claude/CLAUDE.md
: Workspace configuration and guidelines.claude/commands/
: Custom commands as Markdown files
Available commands:
new-requirement
- Complete workflow for adding a new feature from requirements to PRcode-review
- Structured local code review against design docs before pushing changesexecute-plan
- Walk a feature plan task-by-task with interactive promptswriting-test
- Write unit/integration tests targeting 100% coverageupdate-planning
- Update planning and task breakdowncheck-implementation
- Compare implementation with designreview-design
- Review system design and architecturereview-requirements
- Review and summarize requirements
Commands can be referenced in Claude Code chats to guide AI assistance through your development phases.
-
Initialize your project:
ai-devkit init
-
Start with requirements:
- Fill out
docs/ai/requirements/README.md
- Use your AI assistant to help clarify and document requirements
- Fill out
-
Design your system:
- Complete
docs/ai/design/README.md
and feature-specific files - Include mermaid diagrams for architecture, component interactions, and data flow
- Reference requirements when making design decisions
- Complete
-
Plan your work:
- Break down tasks in
docs/ai/planning/README.md
- Estimate and prioritize
- Break down tasks in
-
Implement with guidance:
- Follow patterns in
docs/ai/implementation/README.md
- Keep implementation notes updated
- Follow patterns in
-
Test thoroughly:
- Use
docs/ai/testing/README.md
as your testing guide - Document test cases and results
- Use
-
Deploy confidently:
- Follow deployment procedures in
docs/ai/deployment/README.md
- Follow deployment procedures in
-
Monitor and iterate:
- Set up monitoring per
docs/ai/monitoring/README.md
- Set up monitoring per
Use the /new-requirement
command for a guided workflow:
- In Cursor or Claude Code, type
/new-requirement
- The AI will guide you through:
- π Capturing requirement details
- π Creating feature-specific documentation
- π Designing the solution
- π Planning tasks and breaking down work
- π» Implementation (task by task)
- β Testing and verification
- π Git commits and PR/MR creation
Review and refine your documentation:
- After drafting requirements, run
/review-requirements
to validate completeness - After drafting design, run
/review-design
to ensure architecture clarity and mermaid diagrams
Execute your plan:
- Run
/execute-plan
to step through tasks interactively:- Reads
docs/ai/planning/feature-{name}.md
- Presents tasks in order with context
- Captures status/notes for each task
- Prompts you to update documentation as you progress
- Reads
Before pushing your code:
- Run
/code-review
to perform a structured local review:- Checks alignment with design docs
- Spots logic/security/performance issues
- Highlights redundant code and missing tests
- Suggests documentation updates
Generate comprehensive tests:
- Run
/writing-test
to create unit and integration tests targeting 100% coverage
This workflow creates feature-specific files:
docs/ai/requirements/feature-{name}.md
docs/ai/design/feature-{name}.md
docs/ai/planning/feature-{name}.md
docs/ai/implementation/feature-{name}.md
docs/ai/testing/feature-{name}.md
- New Projects: Scaffold complete development documentation
- Existing Projects: Add structured documentation gradually
- Team Collaboration: Share common development practices
- AI Pair Programming: Provide context for AI assistants
- Knowledge Management: Document decisions and patterns
- Keep templates updated: As your project evolves, update phase documentation
- Reference across phases: Link requirements to design, design to implementation
- Use with AI assistants: Templates are designed to work well with AI code assistants
- Customize for your needs: Templates are starting points, not rigid requirements
- Track decisions: Document architectural decisions and their rationale
The .ai-devkit.json
file tracks your setup:
{
"version": "0.2.0",
"environment": "cursor",
"initializedPhases": ["requirements", "design", "planning"],
"createdAt": "2025-10-14T...",
"updatedAt": "2025-10-14T..."
}
To work on ai-devkit itself:
# Clone the repository
git clone <repository-url>
cd ai-devkit
# Install dependencies
npm install
# Run in development mode
npm run dev init
# Build
npm run build
# Test locally
npm link
ai-devkit init
Note:
ai-devkit init
now ensures the current directory is a git repository. If git is available and the repo isn't initialized, it will rungit init
automatically.
Contributions are welcome! Please feel free to submit issues and pull requests.
MIT
Happy building with AI! π
Task | Command |
---|---|
Initialize everything | npx ai-devkit init --all |
Initialize for Cursor | npx ai-devkit init --environment cursor |
Add specific phases | npx ai-devkit init --phases requirements,design |
Add one phase later | npx ai-devkit phase testing |
Guided feature workflow | /new-requirement (Cursor & Claude) |
Execute feature plan | /execute-plan (Cursor & Claude) |
Generate tests | /writing-test (Cursor & Claude) |
Local code review | /code-review (Cursor & Claude) |
Help | npx ai-devkit --help |
Quick links | Description |
---|---|
CHANGELOG.md | Recent changes and release notes |
templates/ | Phase and environment templates |