init-package

renorm:init-package — Scientific Package Scaffolding

Generate a new scientific computing package in Python, Julia, or C++ from bundled templates.

Workflow

1. Identify language and package name.

   • If the user said "Python package called my_pkg", language is python, name is my_pkg.
   • If the user said "Julia package MyPkg", language is julia, name is MyPkg (PascalCase).
   • If only one is unclear, ask. Do not guess.

2. Identify target directory.

   • Default: a new subdirectory in the current working directory named after the package.
   • User may override with an explicit path.

3. Locate the template directory.

   • For Python: <this-skill>/templates/python/
   • For Julia: <this-skill>/templates/julia/
   • For C++: <this-skill>/templates/cpp/

4. Recursively copy templates to the target directory, applying:

   File content substitutions (in every file, including filenames):

   • {{PACKAGE_NAME}} → user's package name (snake_case for Python/C++, PascalCase for Julia)
   • {{PACKAGE_NAME_PASCAL}} → user's package name in PascalCase (Julia uses this for the module file)
   • {{AUTHOR_NAME}} → from git config user.name
   • {{AUTHOR_EMAIL}} → from git config user.email
   • {{YEAR}} → current year
   • {{DATE}} → current date (YYYY-MM-DD)

   Directory renames:

   • __pkg__ → user's package name (snake_case)
   • __PkgPascal__ → user's package name (PascalCase)

5. Verify the result is well-formed:

   • Python: pyproject.toml parses as TOML.
   • Julia: Project.toml parses as TOML.
   • C++: CMakeLists.txt is present.

6. Report what was created, with the exact files and a one-line summary per file.

7. Do not initialize git, install dependencies, or run tests automatically. Tell the user the next steps:

   • Python: cd <pkg> && pip install -e ".[dev]"
   • Julia: cd <pkg> && julia --project=. -e 'using Pkg; Pkg.instantiate()'
   • C++: cd <pkg> && cmake -B build -S . && cmake --build build

Conventions

The generated packages follow the user's existing scientific-computing rules at ~/.claude/rules/{python,julia,cpp}.md:

• Random number generation is seeded for reproducibility.
• Units are documented in comments (e.g., # Kelvin, # eV).
• Functions have type hints / annotations.
• Tests are scaffolded (pytest for Python, Test stdlib for Julia, simple CMake test for C++).

The skill body does not restate these conventions because they're already in the user's rules and loaded into context via ~/.claude/CLAUDE.md.

When this skill should NOT fire

• The user already has a package and wants to add a feature → just edit existing code.
• The user wants a different language (Rust, Fortran, Lua) → say so and offer to add a template directory.
• The user wants a non-scientific Python project (web app, CLI tool) → this skill is opinionated for scientific computing and may not fit.