Python Sensei — Best Practices Enforcer

Version: 1.1.0 | Author: Shadows Company | License: MIT

WHEN TO TRIGGER

• Writing new Python code
• Reviewing existing Python code
• User says "python review", "best practices", "clean up this python"
• Refactoring Python modules
• Setting up a new Python project

WHEN NOT TO TRIGGER

• Quick scripts where quality doesn't matter
• User explicitly says "just make it work"
• Non-Python code

PREREQUISITES

Requires python or python3 on PATH for syntax checking (python -m py_compile) and test execution (python -m pytest).

Optional tools (auto-detected, recommendations adapt accordingly):

• pytest — test execution (pip install pytest)
• mypy or pyright — static type checking
• ruff — fast linting and formatting (pip install ruff)

The skill checks which tools are available and tailors its recommendations to the user's installed toolchain.

CODE STANDARDS

1. Module Structure

"""Module docstring — one line describing purpose."""

# Standard library imports
import os
from pathlib import Path

# Third-party imports
import httpx
from pydantic import BaseModel

# Local imports
from .config import Settings

# Constants
MAX_RETRIES = 3
DEFAULT_TIMEOUT = 30

# Module code...

Rules:

• Max 500 lines per module — split if larger
• Imports grouped: stdlib → third-party → local (blank line between each group)
• Constants at top, ALL_CAPS naming
• One class per file for complex classes

2. Functions

async def fetch_user(user_id: str, *, include_profile: bool = False) -> User | None:
    """Fetch a user by ID.

    Args:
        user_id: The unique user identifier.
        include_profile: Whether to include full profile data.

    Returns:
        User object if found, None otherwise.

    Raises:
        ConnectionError: If the API is unreachable.
    """

Rules:

• Type hints on all public functions (params + return)
• Keyword-only args after * for clarity
• None return type when failure is normal (not exceptions)
• Docstrings on public functions only
• Max 30 lines per function — extract helpers if larger

3. Data Models

from dataclasses import dataclass, field
from enum import StrEnum

class Status(StrEnum):
    ACTIVE = "active"
    INACTIVE = "inactive"
    PENDING = "pending"

@dataclass
class User:
    id: str
    name: str
    status: Status = Status.ACTIVE
    tags: list[str] = field(default_factory=list)

    def to_dict(self) -> dict:
        return {
            "id": self.id,
            "name": self.name,
            "status": self.status.value,
            "tags": self.tags,
        }

Rules:

• @dataclass for simple models, Pydantic for validation-heavy models
• StrEnum for states (JSON-serializable)
• Always include to_dict() for serialization
• Immutable by default (frozen=True when appropriate)

4. Error Handling

# GOOD: Specific exceptions, minimal try blocks
try:
    response = await client.get(url)
    response.raise_for_status()
except httpx.TimeoutException:
    logger.warning("Request timed out: %s", url)
    return None
except httpx.HTTPStatusError as e:
    logger.error("HTTP %d: %s", e.response.status_code, url)
    raise

# BAD: Catching everything
try:
    do_everything()
except Exception:
    pass  # Never do this

Rules:

• Catch specific exceptions, never bare except:
• Only validate at system boundaries (user input, external APIs)
• Trust internal code — do not add defensive checks everywhere
• Use logging module, not print() for errors

5. Async Patterns

import asyncio
import httpx

async def fetch_all(urls: list[str]) -> list[dict]:
    async with httpx.AsyncClient() as client:
        tasks = [client.get(url) for url in urls]
        responses = await asyncio.gather(*tasks, return_exceptions=True)
        return [r.json() for r in responses if not isinstance(r, Exception)]

Rules:

• async with for resource management
• asyncio.gather() for parallel I/O
• return_exceptions=True to handle partial failures
• Never mix sync and async I/O in the same function

6. Testing

import pytest

class TestUserService:
    def test_create_user_with_valid_data(self):
        user = create_user(name="Alice", email="alice@example.com")
        assert user.name == "Alice"
        assert user.status == Status.ACTIVE

    def test_create_user_rejects_empty_name(self):
        with pytest.raises(ValueError, match="name cannot be empty"):
            create_user(name="", email="alice@example.com")

    @pytest.mark.asyncio
    async def test_fetch_user_returns_none_for_missing(self):
        result = await fetch_user("nonexistent-id")
        assert result is None

Rules:

• Test file: tests/test_{module}.py
• Test names describe the scenario: test_[action]_[condition]_[expected]
• One assertion per test (prefer)
• Use pytest.raises for expected exceptions
• Use fixtures for shared setup

7. Project Setup

project/
  src/
    project_name/
      __init__.py
      main.py
      config.py
      models.py
  tests/
    test_main.py
    test_models.py
  pyproject.toml
  requirements.txt
  .gitignore

pyproject.toml over setup.py. Pin major versions in requirements.txt.

ANTI-PATTERNS TO FLAG

SECURITY CONSIDERATIONS

This skill reads and writes Python source files within the working directory. It does not access files outside the project scope.

• Commands executed: python -m py_compile <file> for syntax checking, python -m pytest <test_file> for test execution. These run local project code — only use on trusted repositories or in sandboxed environments.
• Data read: Source files in the working directory only. No access to secrets, credentials, or system files.
• Network access: None required. The skill operates entirely offline.
• Credentials: None stored or accessed.
• Persistence: All modifications are to source files in the working directory only. No global config changes.
• Sandboxing: Recommended to run in a virtual environment (venv) to isolate dependencies.

OUTPUT FORMAT

Reviews and code output follow the standards defined above. Each review identifies specific anti-patterns with line references and provides corrected code blocks.

RULES

1. Type hints everywhere — public functions must have full type annotations
2. 500 lines max — split large modules into focused sub-modules
3. Test every module — tests/test_{module}.py with descriptive test names
4. Async by default — use async for any I/O operation
5. pathlib over os.path — modern Python path handling
6. No print debugging — use logging module with appropriate levels

Published by Shadows Company — "We work in the shadows to serve the Light."