What happens if CLAUDE.md already exists?

The skill will not automatically overwrite CLAUDE.md. It will either skip generation or present the existing content for review and ask for approval before updating.

What if the project is empty?

The skill prompts you for preferences (stack, conventions, folder layout) to tailor CLAUDE.md and LessonsLearned.md before generating output.

How do I trigger the setup?

Use one of the triggers: setup sdf, iniciar projeto, configurar sdf, or /sdf-setup to activate the process.

sdf-setup

Scanned

npx machina-cli add skill caiobrian/sdf-skills/sdf-setup --openclaw

Files (1)

SKILL.md

1.6 KB

SDF Setup — Project Constitution

Initialize foundational artifacts: .claude/CLAUDE.md and LessonsLearned.md.

Process

Step 1: Analyze the codebase

Examine: stack, framework, bundler, runtime, code patterns, test framework, linters/formatters, folder structure, package scripts.

If the project is empty, ask for preferences.

Step 2: Generate `.claude/CLAUDE.md`

Keep under 60 lines. Do NOT include what linters already enforce.

# Project Constitution

## Stack
- [detected]
- Prohibited: [what NOT to use]

## Tech Decisions
- [e.g., Data fetching: TanStack Query, never fetch + useEffect]

## Code Conventions
- [extracted from codebase]

## Quality Standards
- [from linter/formatter configs]

## Prohibited Actions
- Never install new dependencies without explicit approval

## Folder Structure
- [mapped from project]

## Commands
- Build: [cmd]
- Test: [cmd]
- Lint: [cmd]

Step 3: Generate `LessonsLearned.md`

# Lessons Learned
<!-- Format: LL-XXX: Title | Context | Rule | Date -->

Step 4: Confirm

Present output and ask (in the user's preferred language):

If any important convention is missing
If there are any specific prohibitions

Rules

Detect patterns from REAL codebase, never assume defaults
Always communicate in the user's preferred language

Source

git clone https://github.com/caiobrian/sdf-skills/blob/main/skills/sdf-setup/SKILL.mdView on GitHub

Overview

SDF Setup initializes project constitution by creating CLAUDE.md and LessonsLearned.md for the SDF workflow. It activates when onboarding a new repository, starting a new project, or when no CLAUDE.md exists.

How This Skill Works

The skill analyzes the codebase to detect stack, framework, tooling, folder structure, test setup, and conventions. It then generates .claude/CLAUDE.md capturing key decisions, standards, and folder mappings, followed by a LessonsLearned.md skeleton. If the repository is empty, it prompts for user preferences before proceeding.

When to Use It

Onboarding a new repository into SDF
Starting a new project from scratch
No CLAUDE.md exists in the project
Monorepo or multi-package repo setup
Reinforcing standards before major refactors

Quick Start

Step 1: Analyze the codebase to detect stack, tooling, and folder structure
Step 2: Generate .claude/CLAUDE.md with sections and keep it under 60 lines; omit enforced lint rules
Step 3: Create LessonsLearned.md skeleton and present outputs for your review in your preferred language

Best Practices

Detect patterns directly from the real codebase rather than assumptions
Keep CLAUDE.md concise, target under 60 lines
Do not duplicate existing lint or formatter rules; cite current configs instead
Accurately map folder structure, commands, and build/test scripts
Use the Confirm step to validate conventions with the team (preferred language)

Example Use Cases

Onboarding a new React + Node project
Bootstrapping a Python data processing repo without CLAUDE.md
Initializing a monorepo with multiple packages
Refactoring a legacy project to align with SDF standards
Setting up a microservices project with service boundaries

Frequently Asked Questions

Add this skill to your agents