python-module-boundaries

name	python-module-boundaries
description	Design or review regular Python package and module boundaries with explicit public surfaces, disciplined re-exports, safe imports, and clear internal-module contracts.

Purpose

Choose clear, reviewable Python package and module boundaries without drifting into DDD, framework architecture, or packaging policy.

Trigger / When to use

Use this skill when:

• designing or reviewing a Python package's public import surface
• deciding what should be re-exported through __init__.py
• deciding whether a module is public, internal, too broad, or too side-effectful at import time
• reviewing deep imports, wildcard imports, circular imports, or boundary leaks across subpackages

Do not use this skill when:

• the main task is DDD layering, repository architecture, or framework-specific project structure
• the main task is strict type-hint syntax, API-signature design, or ordinary class internals
• the package is a namespace-package or no-__init__.py layout and the task is mainly about that structure

Inputs

• whether the package is a regular package with __init__.py
• the intended public import paths
• which modules are implementation detail only
• whether imports stay within one local scope or cross a clearer subpackage boundary
• whether the package has import-time setup, compatibility re-exports, or circular dependencies

Process

1. Keep the first draft on regular packages with __init__.py; treat namespace-package or no-__init__.py layouts as out of scope and suggest a regular package when an explicit gateway is needed.
2. Define the package's public API through __init__.py with explicit re-exports and __all__; do not make external callers depend on deep imports into nested internal paths.
3. Treat underscore-prefixed modules and subpackages as non-public contracts to external packages; if outside callers need a stable entrypoint, re-export it through the package gateway instead.
4. Require every public module, even a single-file module, to declare __all__; do not expose underscore-prefixed private names through that export surface.
5. Within one local scope, prefer relative imports for internal collaboration; across clearer subpackage boundaries, prefer absolute imports and do not reach into another subpackage's underscore modules.
6. Keep imports declarative and safe: no wildcard imports, no import-time I/O, connections, environment-driven runtime validation, background tasks, or hidden registration side effects.
7. Treat circular imports as a boundary smell; first extract shared lower-level contracts, then inspect premature __init__.py coupling, then use if TYPE_CHECKING: for type-only cycles, and only use a function-local import as a documented last resort.
8. Keep executable script and CLI modules thin consumers rather than public API providers; reusable logic belongs in regular package modules, and execution should stay behind main() and/or if __name__ == "__main__":.
9. Treat test support code as boundary-managed code too: reusable test helper packages should expose a clean gateway, while individual test_*.py files remain execution-oriented consumers.
10. When a module becomes too broad, mixes unrelated responsibilities, or blends public and internal sections, recommend splitting it; keep deprecated re-exports only as explicit temporary bridges with a removal plan, replacement path, and deprecation signal.

Examples

• Positive: Re-export User from package/__init__.py, declare __all__ in both the gateway and the public module, keep _adapter.py internal, and move import-time setup into an explicit bootstrap function.
• Negative: Make callers import package.internal.models.user.User, rely on from package.x import *, register plugins just by importing a module, or leave a deprecated re-export in place with no warning or removal note.

Outputs

• a review-ready Python module-boundary rule set or review guide
• defaults for package gateways, module exports, import style, and import-time safety
• local examples for public-surface patterns, anti-patterns, edge cases, and split signals

Boundaries

• Do not define DDD layering, framework module layout, or packaging/distribution policy.
• Do not define strict typing syntax, API-signature rules, or class-internal design rules.
• Do not treat namespace packages as first-draft in-scope; only flag them and hand off.
• Do not claim approved or stable.

Local references

• examples.md: gateway patterns, internal-module rules, import-side-effect anti-patterns, circular-import fixes, script/test edge cases, and deprecated re-export examples