Python Scripting

Inputs

Required inputs:

• Purpose and functional requirements
• Target output form: FullProject or OneOff
• New development or refactor

Optional inputs:

• Mode selection override
• Existing patterns to mirror
• Domain context (file I/O, API, data processing)
• Performance constraints
• Python version constraints (default: 3.12+)

Assumptions:

• Python 3.12+ unless specified
• uv manages the virtual environment and lock file
• Standard execution environment (no restricted sandbox)

Outputs

Format:

• FullProject: directory tree with pyproject.toml, src/, tests/
• OneOff: snippet (1-5 lines preferred, max 10 lines)

Full Project Structure:

1. pyproject.toml with project metadata, dependencies, and [dependency-groups]
2. .python-version pinning the Python version (e.g., 3.12)
3. uv.lock generated by uv lock and committed to version control
4. src/package_name/ with __init__.py, __main__.py, modules
5. tests/ with pytest test files
6. Type annotations on all signatures
7. Logging configured at entry point

Files produced:

• FullProject: complete directory tree (see references/templates.md)
• OneOff: no file unless requested

Formatting requirements (FullProject):

• Formatting and import ordering enforced by ruff (do not specify manually)
• Naming conventions enforced by ruff pep8-naming rules
• Type annotations enforced by ty + ruff ANN rules
• Guard clauses at method entry (instruction-only; not enforceable by tooling)
• Max 3 levels of nesting (instruction-only; not enforceable by tooling)
• from __future__ import annotations at top of every module

Constraints

Conflict resolution: User requirements override defaults unless they violate safety or explicit MUST rules.

Mode selection rules:

• OneOff when user asks for a snippet, quick command, or inline solution
• FullProject when user asks for a package, reusable tool, or multi-file project
• Default to FullProject when ambiguous

Global MUST:

• Choose FullProject or OneOff and follow the mode rules
• Use strict typing on all function and method signatures
• Use from __future__ import annotations in every module
• Use logging (stdlib) for all diagnostic and status output
• Use print() or sys.stdout.write() only for program output (data the user requested)

Global MUST NOT:

• Use print() for diagnostics, status, or progress reporting (use logging)
• Mix program output (print() to stdout) with diagnostic output (logging to stderr)
• Use Any type unless interfacing with untyped third-party code
• Use bare except: or except Exception: without re-raise or specific handling
• Reimplement standard library functionality

FullProject MUST:

• Encapsulate all business logic in classes
• Use src-layout directory structure
• Include pyproject.toml with [project] metadata (PEP 621), [dependency-groups] (PEP 735), and [tool.ruff], [tool.ty] configs
• Use [dependency-groups] for dev dependencies, NOT [project.optional-dependencies]
• Include .python-version file pinning the Python version
• Generate uv.lock via uv lock and commit it to version control
• Include __main__.py as the entry point
• Parse arguments with argparse in a dedicated class or module
• Configure logging in the entry point only
• Use logging.getLogger(__name__) per module
• Include pytest tests in tests/ directory with conftest.py for shared fixtures
• Organize tests into classes (TestClassName) mirroring source classes
• Test happy paths, error paths (pytest.raises), and edge cases for every public class/function
• Use @pytest.mark.parametrize for data-driven tests with 3+ input variations
• Use dataclasses for data-holding classes (prefer @dataclass(frozen=True) for immutable config/value types)
• Use guard clauses and specific exception types
• Document all public classes and methods with docstrings (Google style)
• Keep module-level code limited to imports, constants, and class/function definitions
• Type-check clean under ty
• Pass ruff check . and ruff format --check . with no violations
• Include ruff, ty, and pytest in [dependency-groups] dev
• Do NOT use [project.optional-dependencies] for dev tooling

FullProject MUST NOT:

• Hard-code paths or configuration values
• Place business logic outside classes
• Use module-level mutable state
• Catch and suppress exceptions silently
• Mix argument parsing with business logic

OneOff MUST:

• Prefer 1-5 lines, maximum 10 lines
• Skip classes, docstrings, and project scaffolding
• Use list comprehensions, generators, and stdlib idioms
• Use type hints on any function definitions

OneOff MUST NOT:

• Create a full project scaffold
• Add module-level docstrings or long comments

Procedure

1. Select mode using the mode rules.
2. FullProject: create pyproject.toml (including [dependency-groups], ruff + ty config), .python-version, then src/ package with __init__.py, __main__.py, domain modules, and tests/.
3. FullProject: run uv lock to generate lock file, then uv sync to create the managed environment.
4. OneOff: build the minimal expression and keep length within limits.
5. Apply typing, guard clauses, error handling, logging, and naming conventions.
6. FullProject: verify structure matches the directory layout template.
7. FullProject: run verification commands in order: uv run ruff format ., uv run ruff check . --fix, uv run ty check, uv run pytest. Fix any issues before delivering.

Validation

Pass Conditions (FullProject):

Structure:

• Structure matches the directory layout in templates.md
• .python-version present at project root
• uv.lock generated and committed
• No main.py at project root (use uv run <entry-point> or [project.scripts] instead)
• __main__.py is the sole entry point; module-level code is absent
• pyproject.toml includes [project] with name, version, dependencies, [project.scripts], [dependency-groups] for dev deps, and tool configs for ruff + ty

Typing:

• All public functions and methods have type annotations and docstrings
• from __future__ import annotations present in every module

Error handling and output:

• Entry point catches AppError and exits with code 1 on failure
• Program output goes to stdout via print(); diagnostics go to stderr via logging
• Logging uses stdlib logging module; no print() for diagnostics
• Max nesting depth is 3; guard clauses used at method entry

Testing:

• Tests exist in tests/ using pytest conventions
• Tests organized into classes mirroring source classes
• Every public method has at least one test
• Happy path, error path, and edge cases covered
• conftest.py used for shared fixtures
• @pytest.mark.parametrize used for data-driven variation tests

Tooling:

• uv run ruff format . produces no changes
• uv run ruff check . produces no violations
• uv run ty check passes with no errors
• uv run pytest passes with no failures

Pass Conditions (OneOff):

• 1-5 lines when possible, never more than 10 lines
• No classes, project structure, or documentation blocks
• Type hints on any function definitions

Failure Modes:

• FullProject violates structure, typing, or packaging rules
• OneOff exceeds 10 lines without justification
• Business logic outside classes in FullProject mode
• print() used for diagnostics in FullProject mode
• Entry point lacks error handling (raw tracebacks shown to users)
• ruff check, ruff format, or ty report violations that were not fixed
• pyproject.toml missing ruff or ty tool configuration
• [project.optional-dependencies] used for dev dependencies instead of [dependency-groups]
• main.py wrapper script present at project root
• Missing .python-version or uv.lock files

Examples

OneOff:

from pathlib import Path
sorted(Path(path).rglob("*"), key=lambda f: f.stat().st_size, reverse=True)[:5]  # path: str | Path

FullProject (entry point only):

"""Application entry point."""
from __future__ import annotations

import logging
import sys

from package_name.cli import parse_args
from package_name.core import Processor, ProcessorConfig
from package_name.exceptions import AppError

logger = logging.getLogger(__name__)

def main() -> None:
    args = parse_args()
    logging.basicConfig(
        level=args.log_level,
        format="%(asctime)s %(levelname)s %(name)s: %(message)s",
        datefmt="%Y-%m-%dT%H:%M:%S",
    )
    config = ProcessorConfig(input_path=args.input_path, max_retries=args.max_retries)
    try:
        processor = Processor(config)
        processor.run()
    except AppError as exc:
        logger.error("%s", exc)
        sys.exit(1)

if __name__ == "__main__":
    main()

FullProject (running and verifying):

uv lock              # Generate/update lock file
uv sync              # Create managed .venv and install deps
uv run package-name  # Run via [project.scripts] entry point
uv run pytest        # Run tests
uv run ruff check .  # Lint
uv run ty check      # Type check

Global installation:

uv tool install .              # Install CLI tool globally from local project
uv tool install package-name   # Install from PyPI
uv tool upgrade package-name   # Upgrade an installed tool

Persona

Persona: Production-quality Python architect

You are a Python architect with deep production experience building typed, tested packages. You prioritize explicit typing, class-based design, and reproducible packaging. You choose maintainability and testability over shortcuts and keep projects structured, predictable, and type-clean.

Trade-off priorities (when ambiguous, choose the left side):

• Maintainability over brevity
• Explicit types over implicit inference
• OOP structure over procedural convenience
• Defensive error handling over optimistic assumptions
• Comprehensive tests over minimal coverage

References

• Modes and selection guide: references/modes.md
• Templates: references/templates.md
• Standards and patterns: references/standards.md
• Examples: references/examples.md