AgentSkillsCN

prnd-layered-python-code-reviewer

在代码审查时，重点关注分层架构边界、依赖注入与依赖倒置原则、配置验证、类型标注的质量（mypy/flake8 准备情况）、命名与文档字符串的准确性，以及可测试性。当被要求进行“代码审查”“PR 审查”，或简单地表达“帮我看看吧”时，此技能有助于避免因架构回归或低价值改动而反复反馈。

SKILL.md
--- frontmatter
name: prnd-layered-python-code-reviewer
description: Review Python code changes with a focus on layered architecture boundaries, DI/DIP, configuration validation, typing quality (mypy/flake8 readiness), naming/docstring accuracy, and testability. Use when asked to “code review”, “PR review”, or “리뷰해줘” to prevent repeat feedback on architecture regressions and low-signal changes.

Layered Python Code Reviewer

Overview

Provide a reusable code review workflow and checklist that applies across Python projects, especially ones that aim for layered architecture and dependency inversion. Produce actionable, concrete review comments (not vague style opinions).

Review Workflow

  1. •

    Identify the change scope

    • •Summarize touched modules and layers (example layers: domain/, service/, dataset/, adapters/, infra/).
    • •Identify which changes are: (a) core policy/logic, (b) boundary I/O, (c) data shaping, (d) test-only.
  2. •

    Apply hard blockers first

    • •Dependency direction (DIP): low-level modules must not import high-level policy/service types.
    • •Public interfaces: avoid leaking raw dict across layers; use TypedDict, dataclasses, or Pydantic models.
    • •Config correctness: validate invariants on load (range continuity, required keys, value bounds).
    • •Type/lint hygiene: fix None handling, unused variables, and any Any propagation caused by missing types.
  3. •

    Apply design-quality checks

    • •Composition root: wiring (ENV/config/I/O/client construction) stays at the boundary (create()/entrypoint).
    • •Separation of concerns: “build payload/message” vs “send request”, “decide” vs “orchestrate”.
    • •Test seams: time/randomness/I/O are injectable; core logic is deterministic.
  4. •

    Write review output

    • •Group comments by: Architecture, DIP/DI, Typing, Config, Error handling, Tests, Naming/Docs.
    • •For each issue: state impact + preferred pattern + minimal change suggestion.

References

Load and apply:

  • •references/principles.md for a project-agnostic checklist and comment templates.
  • •references/review-themes.md for common “review-driven refactor” patterns (kept generic on purpose).

Optional Automation

If your project uses clear “layer folders” under a package root, run:

python scripts/check_layered_imports.py --root <package_root_dir> --package-name <top_level_package>

Use flags to match your project layout:

  • •--layers domain,service,adapters,infra
  • •--forbid domain:service,adapters,infra
  • •--forbid adapters:service