Flask Expert guides Claude through Flask 3.1 work in five steps: structure the app with a factory, model data with typed SQLAlchemy 2.x, build a JSON or Jinja interface, secure it and test and deploy it. New apps get the application factory pattern and settings read from FLASK_* environment variables, and the app refuses to start without a secret key.
The skill also carries a fourteen-item audit for existing projects. Each item is a grep command with the result it should give, such as no db.Column, no Model.query, no hardcoded secret and no debug=True. Seven reference files cover structure, database, REST APIs, templates and forms, authentication, testing and deployment.
What You Get
- An application factory with blueprints, extensions and environment-based configuration
- Typed Flask-SQLAlchemy models, select() queries and Flask-Migrate migrations
- JSON blueprints with Pydantic or marshmallow validation and an app-level JSON error handler
- Flask-Login or Flask-JWT-Extended authentication with ownership checks
- pytest fixtures over the factory and gunicorn or waitress deployment with Celery, RQ or APScheduler jobs
- A fourteen-step grep audit of an existing Flask project
Common Use Cases
- Starting a Flask JSON API with validated bodies and migrations
- Adding login, forms with CSRF or JWT tokens to an existing Flask app
- Modernizing a legacy project that uses db.Column, Model.query and a module-level app
- Fixing errors such as working outside of application context or a 500 that only shows under gunicorn
- Preparing a Flask app for production behind a reverse proxy
What Makes It Unique
Most framework skills describe how to write new code. This one adds an ordered audit of an existing project, with a grep for each common mistake and the answer it should return, and it says to keep an existing app’s structure and suggest the factory without converting it unasked. It also covers the Flask 3.x removals, Flask-SQLAlchemy 2.x pins, and the gap between SQLite tests and PostgreSQL.
Under The Hood
The skill is a SKILL.md with seven reference files and no scripts; the references load by topic. The agent is told to run grep commands over the project, run the project’s pytest suite before calling work done, and run flask db migrate and flask db upgrade, so it uses the shell and writes code, tests and migration files. It also reads git state to check that .env is ignored. Examples read secrets from environment variables, hash passwords and issue session or JWT tokens. The skill does not fetch web pages or call outside APIs.