| name | project-init |
| description | Automatically detects new project initialization, collaborates with user on project planning, and sets up the appropriate tech stack with matching skills and agents. Use when starting a new project, creating a new repository, or working in an empty/minimal directory that needs project structure. |
| type | domain |
| enforcement | suggest |
| priority | high |
| keywords | new project, start project, initialize project, project setup, empty directory, tech stack |
| patterns | (start|create|init|setup|initialize).*?(project|repo|repository), new.*?(project|repo) |
Project Initialization
Overview
This skill automates project setup through collaborative planning. It detects when you're starting a new project, identifies the tech stack, presents skill and agent recommendations, collaborates with the user on the project plan, then initializes the appropriate project structure with matching skills and agents.
When to Activate
Activate this skill when:
- Working in an empty or near-empty directory
- User explicitly mentions "new project", "start a project", or "initialize"
- No existing project files detected (go.mod, package.json, Cargo.toml, pyproject.toml, etc.)
- User asks to set up a specific tech stack
Do NOT activate for:
- Existing projects with established structure
- Single-file scripts or experiments
- Documentation-only work
Workflow
Phase 1: Detection
Check for new project indicators:
-
Directory Analysis
ls -la
find . -maxdepth 2 -name "go.mod" -o -name "package.json" -o -name "Cargo.toml" -o -name "pyproject.toml" -o -name "pom.xml"
-
Detection Criteria
- Empty directory OR
- Only contains: README.md, .git, .gitignore, LICENSE OR
- User explicitly requests new project setup
Phase 2: Stack Identification
Determine the tech stack through:
- Explicit user statement - "create a Go project", "new Node.js app"
- Context clues - user mentions specific frameworks, tools, or languages
- Ask the user if unclear
Common stacks:
- Go - Backend services, CLI tools
- Node.js - Web apps, APIs, full-stack
- Python - Data science, ML, scripting, web apps
- Rust - Systems programming, performance-critical apps
- TypeScript - Frontend, full-stack web apps
Phase 3: Planning & Skill Recommendations
CRITICAL: Before initializing anything, collaborate with the user on the project plan.
-
Present stack-specific recommendations
Based on identified stack, suggest:
- Core skills - Essential for the stack (e.g., go-tdd for Go)
- Workflow skills - Development workflow automation (linting, formatting, CI/CD)
- Domain-specific skills - Based on project type (web-api, cli-tool, data-pipeline, etc.)
- Agent suggestions - Specialized agents that could help (testing agent, documentation agent, etc.)
-
Stack-specific recommendations:
Go Projects:
- Skills:
go-tdd (required), go-lint (golangci-lint), go-api (REST API patterns)
- Agents:
go-test-agent (runs tests on file changes), go-doc-agent (generates godoc)
- Tools: Consider air (hot reload), sqlc (type-safe SQL), wire (dependency injection)
Node.js/TypeScript:
- Skills:
node-tdd (Jest/Vitest), typescript-lint (ESLint/Prettier), node-api (Express/Fastify patterns)
- Agents:
test-watch-agent, bundle-analyzer-agent
- Tools: Consider tsx (TypeScript runner), vitest (fast testing), tsup (bundler)
Python:
- Skills:
python-tdd (pytest), python-lint (black/ruff/mypy), python-api (FastAPI patterns)
- Agents:
pytest-watch-agent, coverage-reporter-agent
- Tools: Consider poetry (dependency mgmt), ruff (fast linter), mypy (type checking)
Rust:
- Skills:
rust-tdd, rust-clippy, rust-api (axum/actix patterns)
- Agents:
cargo-watch-agent, clippy-agent
- Tools: Consider cargo-watch, cargo-audit, sccache (compiler cache)
-
Ask user for preferences using AskUserQuestion tool
Present 2-4 questions about:
- What type of project? (API, CLI, library, full-stack app, etc.)
- Which skills to create/activate?
- Agent personality preference? (Friendly, Professional, Analytical, Custom)
- Which additional agents? (test-runner, linter, security-auditor, etc.)
- Development workflow preferences? (watch mode, pre-commit hooks, CI/CD)
-
Create a plan summary
📋 Project Plan: <project-name>
Stack: <language/framework>
Type: <project-type>
Skills to create:
- <skill-1> - <purpose>
- <skill-2> - <purpose>
Agents to consider:
- <agent-1> - <purpose>
- <agent-2> - <purpose>
Project structure:
- <directory-tree>
Next steps:
1. Initialize project structure
2. Create/activate skills
3. Set up tooling
-
Get explicit approval before proceeding
- Wait for user confirmation
- Allow modifications to the plan
- Only proceed when user approves
Phase 4: Project Initialization
Initialize the project based on identified stack:
Go Projects
go mod init <module-name>
mkdir -p cmd pkg internal
cat > .gitignore << 'EOF'
*.exe
*.dll
*.so
*.dylib
bin/
dist/
*.out
coverage.html
.idea/
.vscode/
*.swp
*.swo
*~
.DS_Store
Thumbs.db
EOF
Skills to create/activate:
go-tdd - Test-driven development for Go
Node.js/TypeScript Projects
npm init -y
npm install -D typescript @types/node
npx tsc --init
mkdir -p src tests
npx gitignore node
Skills to create/activate:
node-tdd - Test-driven development with Jest/Vitest
typescript-lint - TypeScript linting and formatting
Python Projects
cat > pyproject.toml << 'EOF'
[build-system]
requires = ["setuptools>=61.0"]
build-backend = "setuptools.build_meta"
[project]
name = "project-name"
version = "0.1.0"
description = ""
requires-python = ">=3.8"
dependencies = []
[project.optional-dependencies]
dev = ["pytest>=7.0", "black", "ruff"]
EOF
mkdir -p src tests
cat > .gitignore << 'EOF'
__pycache__/
*.py[cod]
*$py.class
*.so
.Python
venv/
.env
.pytest_cache/
.coverage
htmlcov/
dist/
build/
*.egg-info/
EOF
Skills to create/activate:
python-tdd - Test-driven development with pytest
python-lint - Black, ruff, mypy integration
Rust Projects
cargo init
Skills to create/activate:
rust-tdd - Test-driven development with Rust
rust-clippy - Clippy integration
Phase 5: Skills and Agents Setup
After initializing the project structure, create recommended skills and agents.
Creating Skills
-
Check existing skills
ls .claude/skills/
-
For each recommended skill:
If skill exists (e.g., go-tdd):
- Mention it's already available
- Explain how to activate/use it
If skill doesn't exist:
-
Collaborative approach:
- Ask user which skills they want to create
- Create them one by one
- Customize each based on user preferences
- Don't use static templates - build dynamically with user
Creating Agents
Create agents using agent-creator guidance (no scripts needed - just Write tool).
ALWAYS create orchestrator agent first - required for every project.
1. Create Orchestrator (Required)
The orchestrator coordinates all work and delegates to specialists. Customize nickname based on user's personality preference:
Nickname suggestions:
- Friendly: Buddy, Pal, Coach (^_^)
- Professional: Maestro, Lead, Coordinator (•_•)
- Analytical: Architect, Planner, System (⊙_⊙)
- Quirky: Captain, Chief, Boss (⌐■_■)
---
name: orchestrator
description: (<emoticon>) <Nickname> - The task coordinator who gathers info and delegates execution
model: sonnet
color: cyan
---
You are <Nickname>, the project orchestrator.
[Customize prompt based on personality: Friendly, Professional, Analytical, Quirky]
- Core behavior: Gather context, never execute changes, always delegate
- Communication style: Match user's preference
2. Create Stack-Specific Agents
For each additional recommended agent:
-
Determine agent details collaboratively:
- Name (e.g.,
go-test-runner)
- Description with trigger keywords
- Tools needed (minimal access)
- Model choice (haiku for simple, sonnet for complex)
- Color for visual distinction
- Personality matching user preference
-
Create agent file at .claude/agents/<agent-name>.md:
---
name: <agent-name>
description: <when and why to use this agent>
model: <haiku|sonnet|opus>
color: <blue|green|red|yellow|purple|orange|pink|gray|cyan>
tools: <optional comma-separated list>
---
<System prompt with personality applied>
-
Stack-specific agent examples:
Go projects:
go-test-runner (green, haiku) - Runs go test on changes
go-linter (yellow, haiku) - Runs golangci-lint
code-reviewer (blue, sonnet) - Reviews Go code quality
Node.js/TypeScript:
vitest-runner (green, haiku) - Runs Vitest tests
type-checker (yellow, haiku) - Runs tsc --noEmit
bundle-analyzer (yellow, sonnet) - Analyzes bundle size
Python:
pytest-runner (green, haiku) - Runs pytest with coverage
ruff-linter (yellow, haiku) - Runs ruff linter
mypy-checker (yellow, haiku) - Type checking with mypy
Rust:
cargo-test-runner (green, haiku) - Runs cargo test
clippy-linter (yellow, haiku) - Runs clippy
cargo-auditor (red, sonnet) - Security audit
-
Agent creation is fully collaborative:
- Ask user which agents to create
- Customize descriptions and prompts together
- No static templates - build what user needs
- Reference agent-creator skill for template guidance
Phase 6: Confirmation
Provide user with summary:
✅ Project initialized: <stack-name>
✅ Created: <list of files>
✅ Activated skills: <list of skills>
Next steps:
- <stack-specific getting started tips>
Stack-Specific Templates
Go Project Structure
.
├── go.mod
├── .gitignore
├── README.md
├── cmd/
│ └── app/
│ └── main.go
├── pkg/
│ └── (shared libraries)
└── internal/
└── (private code)
Node.js/TypeScript Structure
.
├── package.json
├── tsconfig.json
├── .gitignore
├── README.md
├── src/
│ └── index.ts
└── tests/
└── index.test.ts
Python Structure
.
├── pyproject.toml
├── .gitignore
├── README.md
├── src/
│ └── package_name/
│ └── __init__.py
└── tests/
└── test_package.py
Rust Structure
.
├── Cargo.toml
├── .gitignore
├── README.md
└── src/
└── main.rs
Decision Tree
User mentions new project OR directory is empty/minimal
↓
Is stack explicitly stated?
├─ YES → Proceed to planning phase
└─ NO → Can stack be inferred from context?
├─ YES → Confirm with user, then plan
└─ NO → Ask user about stack preference
↓
Present skill & agent recommendations for stack
↓
Ask user preferences (project type, skills, agents, tooling)
↓
Create plan summary
↓
Get user approval for plan
↓
Initialize project structure
↓
Create/activate skills
↓
Create agents if requested
↓
Confirm completion with user
Best Practices
- Always confirm stack choice if there's any ambiguity
- Present recommendations before acting - don't initialize until user approves the plan
- Use AskUserQuestion tool for gathering preferences during planning phase
- Use standard conventions for each ecosystem (don't invent new patterns)
- Create minimal viable structure - avoid over-scaffolding
- Collaborate on skills and agents - build dynamically with user, not from static templates
- Reference skill-creator and agent-creator - use them as guidance for creating skills/agents
- Document agent usage in project README if agents are created
- Provide clear next steps after initialization
Integration with Other Skills
This skill works together with:
Workflow integration:
- project-init detects empty directory
- Ask user about stack and project type
- Suggest relevant skills and agents
- Use skill-creator to scaffold skills if needed
- Use agent-creator patterns to create agents
- Customize everything collaboratively
Common Patterns
Multi-language Projects
For projects with multiple languages (e.g., Go backend + TypeScript frontend):
- Create subdirectories:
backend/, frontend/
- Initialize each with appropriate stack
- Create/activate skills for both stacks
- Add root-level README explaining structure
Monorepo
For monorepo setups:
- Ask about workspace structure (Nx, Turborepo, etc.)
- Initialize workspace root
- Create example packages/apps
- Set up shared tooling
Resources
This skill doesn't require bundled resources - all initialization is done through direct commands and file generation.