Sam Dumont Claude Skills Claude Skills

Advanced Python Code Quality

Goes beyond the basic stack (ruff, mypy, xenon, vulture, pre-commit) to cover every remaining quality dimension. Each section maps a gap to specific tools with exact configuration and thresholds.

Core insight: AI-generated code doesn't need different tools — it needs tighter thresholds on existing tools.

Quick Reference: Tool Inventory

1. Cognitive Complexity

Cyclomatic complexity (xenon) counts execution paths. Cognitive complexity measures how hard code is for a human to understand — deeply nested loops score much higher than flat switch-style logic even at identical cyclomatic complexity.

complexipy (Rust, fast, pre-commit ready)

# pyproject.toml
[tool.complexipy]
paths = ["src"]
max-complexity-allowed = 10
exclude = ["src/generated/**"]

# .pre-commit-config.yaml
- repo: https://github.com/rohaquinlop/complexipy-pre-commit
  rev: v4.2.0
  hooks:
    - id: complexipy

radon Maintainability Index (already installed via xenon)

Gate at rank A (MI ≥ 20) for new code. Halstead metrics flag functions where estimated bugs exceed 0.5.

wily — complexity trend tracking

# GitHub Actions — fail PR if complexity regresses
- name: Check complexity trend
  run: |
    pip install wily
    wily build src/ -n 20
    wily diff src/ -r origin/${{ github.event.pull_request.base.ref }}

2. Code Duplication

Neither Ruff nor mypy detect copy-pasted code blocks.

jscpd (recommended for CI)

Token-based (Rabin-Karp), gates on percentage threshold. Exits non-zero when exceeded.

{
  "threshold": 5,
  "minLines": 5,
  "minTokens": 50,
  "mode": "strict",
  "format": ["python"],
  "ignore": ["**/tests/**", "**/migrations/**"],
  "reporters": ["console", "json"],
  "gitignore": true
}

Alternatives

Pylint R0801: min-similarity-lines = 6, ignore-imports = true. Cannot be suppressed inline — whole-project check.

3. Architectural Enforcement

import-linter (declarative contracts)

Three contract types: layers (one-way deps), independence (zero cross-imports), forbidden (block specific imports).

# pyproject.toml
[tool.importlinter]
root_package = "myapp"
include_external_packages = true
exclude_type_checking_imports = true

[[tool.importlinter.contracts]]
id = "clean-arch"
name = "Clean Architecture layers"
type = "layers"
layers = [
    "myapp.infrastructure",
    "myapp.interface",
    "myapp.application",
    "myapp.domain",
]

[[tool.importlinter.contracts]]
id = "domain-purity"
name = "Domain has no framework dependencies"
type = "forbidden"
source_modules = ["myapp.domain"]
forbidden_modules = ["sqlalchemy", "fastapi", "requests", "redis"]

[[tool.importlinter.contracts]]
id = "bounded-contexts"
name = "Bounded contexts are independent"
type = "independence"
modules = ["myapp.domain.orders", "myapp.domain.users", "myapp.domain.payments"]

# .pre-commit-config.yaml
- repo: https://github.com/seddonym/import-linter
  rev: v2.7
  hooks:
    - id: import-linter

pytestarch — architecture rules as tests

from pytestarch import Rule, EvaluableArchitecture

def test_domain_isolation(evaluable: EvaluableArchitecture):
    rule = (
        Rule()
        .modules_that()
        .are_sub_modules_of("src.myapp.domain")
        .should_not()
        .import_modules_that()
        .are_sub_modules_of("src.myapp.infrastructure")
    )
    rule.assert_applies(evaluable)

pytest-archon alternative: archrule("domain").match("myapp.domain*").should_not_import("myapp.infrastructure*").check("myapp")

4. Dependency Hygiene

deptry (Rust-powered)

Cross-references imports against declared dependencies:

• DEP001: Import of undeclared package
• DEP002: Declared but never imported (bloat)
• DEP003: Transitive dependency used without declaring
• DEP004: Dev dependency used in production code

# pyproject.toml
[tool.deptry]
exclude = ["tests", "docs", "migrations"]

[tool.deptry.per_rule_ignores]
DEP002 = ["celery", "gunicorn"]  # Runtime-only, not imported
DEP004 = ["pytest"]

# .pre-commit-config.yaml
- repo: https://github.com/fpgmaas/deptry.git
  rev: "0.24.0"
  hooks:
    - id: deptry

pipdeptree — dependency conflicts

pipdeptree --warn fail  # Catch conflicting versions

5. Test Quality Gates

diff-cover — PR-level coverage gating

New code should have near-complete coverage even if the overall codebase doesn't.

# GitHub Actions
- run: pytest --cov=src --cov-branch --cov-report=xml --cov-fail-under=80
- run: diff-cover coverage.xml --compare-branch=origin/main --fail-under=95

mutmut — mutation testing

Mutates source code (flips operators, changes constants, modifies returns) and reruns tests. Surviving mutants = weak tests.

mutmut run --paths-to-mutate src/core/
# Gate at ≥75% mutation score for critical modules

mutmut 3+ uses fork() for speed. Set mutate_only_covered_lines = true to skip untested code.

Additional pytest plugins

• pytest-benchmark: --benchmark-compare-fail=mean:10% fails CI on performance regression
• pytest-randomly: Randomizes test order to catch hidden inter-test dependencies
• pytest-deadfixtures: Detects unused fixtures
• hypothesis: Property-based testing — systematically generates edge cases

6. Pylint Design Checks & Refurb

Ruff implements ~209 of pylint's ~409 rules. Missing: type-inference checks, cross-module analysis, design checker suite.

Pylint design thresholds (tighter than defaults)

[tool.pylint.design]
max-args = 6              # Default: 5
max-positional-arguments = 4
max-locals = 12           # Default: 15
max-returns = 5           # Default: 6
max-branches = 10         # Default: 12
max-statements = 40       # Default: 50
max-bool-expr = 4         # Default: 5
max-parents = 5           # Default: 7

refurb — modernization beyond Ruff

Catches idioms Ruff hasn't ported: Path(x).read_text() over open(x).read(), tuple membership over chained or, list.clear() over del x[:].

# .pre-commit-config.yaml
- repo: https://github.com/dosisod/refurb
  rev: v2.0.0
  hooks:
    - id: refurb

fixit (Meta) — custom lint rules with auto-fixes

Built on LibCST. Use for project-specific codemods: enforcing API patterns, migrating deprecated internal calls.

SonarQube/SonarCloud

Comprehensive platform: cognitive complexity + duplication + tech debt + framework-aware rules (Django, FastAPI, Flask, NumPy, Pandas). Imports reports from Ruff, pylint, Bandit, mypy. Free for OSS via SonarCloud.

7. Documentation & Type Coverage

interrogate — docstring coverage

# pyproject.toml
[tool.interrogate]
ignore-init-method = true
ignore-magic = true
ignore-module = true
fail-under = 95
style = "google"
exclude = ["tests", "docs"]

# .pre-commit-config.yaml
- repo: https://github.com/econchick/interrogate
  rev: 1.7.0
  hooks:
    - id: interrogate
      args: [--quiet, --fail-under=95]

Ruff docstring rules

• D rules: Docstring style enforcement (replaces deprecated pydocstyle)
• DOC rules (preview): Docstring-to-signature consistency. Enable DOC202, DOC402, DOC403, DOC501 with preview = true

griffe — API breaking change detection

griffe check mypackage --against pypi  # Block releases with accidental API breaks

Type coverage

• pyright: Faster than mypy, --verifytypes for type completeness
• mypy --any-exprs-report: Percentage of expressions with known types vs Any. Gate at 85%
• MonkeyType: Runtime tracing to bootstrap annotations — monkeytype run -m pytest then monkeytype apply module_name

8. AI-Generated Code Detection

Research: 5.2% of commercial LLM outputs contain hallucinated package imports (USENIX Security 2025, 576K samples). Code cloning rose from 8.3% to 12.3% in AI-heavy codebases (GitClear).

sloppylint

Detects hallucinated imports (4-layer resolution: builtins → stdlib → installed → importlib), cross-language leakage (.push() in Python), placeholder code, mutable defaults.

pip install sloppylint
sloppylint src/

Tightened thresholds for AI-heavy codebases

Compose existing tools with stricter settings:

• complexipy threshold 8 (half the default) — catches nested-logic tendency
• jscpd threshold 3% — catches copy-paste patterns
• pylint max-statements=30 — catches bloated functions
• Ruff B006 — mutable default arguments (top AI mistake)
• Ruff SIM — overly complex patterns AI generates
• Vulture — dead code AI generates but never calls
• deptry DEP001 — imports of undeclared packages
• pip-audit — slopsquatting attacks via hallucinated packages

9. Integration Pattern

Pre-commit: staged by speed

# .pre-commit-config.yaml
default_install_hook_types: [pre-commit, commit-msg, pre-push]
fail_fast: true

repos:
  # ~0.1s — Formatting
  - repo: https://github.com/astral-sh/ruff-pre-commit
    rev: v0.15.6
    hooks:
      - id: ruff-check
        args: [--fix, --exit-non-zero-on-fix]
      - id: ruff-format

  # ~0.2s — Quick validation
  - repo: https://github.com/pre-commit/pre-commit-hooks
    rev: v4.6.0
    hooks:
      - id: check-yaml
      - id: check-toml
      - id: check-added-large-files
      - id: debug-statements

  # ~0.3s — Cognitive complexity
  - repo: https://github.com/rohaquinlop/complexipy-pre-commit
    rev: v4.2.0
    hooks:
      - id: complexipy

  # ~0.3s — Docstring coverage
  - repo: https://github.com/econchick/interrogate
    rev: 1.7.0
    hooks:
      - id: interrogate
        args: [--quiet, --fail-under=95]

  # ~0.5s — Dependency hygiene
  - repo: https://github.com/fpgmaas/deptry.git
    rev: "0.24.0"
    hooks:
      - id: deptry

  # ~0.5s — Architectural boundaries
  - repo: https://github.com/seddonym/import-linter
    rev: v2.7
    hooks:
      - id: import-linter

  # ~0.5s — Secrets
  - repo: https://github.com/gitleaks/gitleaks
    rev: v8.22.1
    hooks:
      - id: gitleaks

  # ~0.5s — Modernization
  - repo: https://github.com/dosisod/refurb
    rev: v2.0.0
    hooks:
      - id: refurb

  # ~2-10s — Type checking (pre-push only)
  - repo: https://github.com/pre-commit/mirrors-mypy
    rev: v1.14.0
    hooks:
      - id: mypy
        stages: [pre-push]
        require_serial: true
        pass_filenames: false
        additional_dependencies: [types-requests, types-PyYAML]

  # Conventional commits
  - repo: https://github.com/compilerla/conventional-pre-commit
    rev: v3.6.0
    hooks:
      - id: conventional-pre-commit
        stages: [commit-msg]

CI: parallel jobs with quality gate

jobs:
  pre-commit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
      - uses: pre-commit/action@v3.0.1

  typecheck:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - run: pip install -e ".[dev]"
      - run: mypy --strict src/

  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with: { fetch-depth: 0 }
      - run: pip install -e ".[dev]"
      - run: pytest --cov=src --cov-branch --cov-report=xml --cov-fail-under=80
      - run: diff-cover coverage.xml --compare-branch=origin/main --fail-under=95

  duplication:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - run: npx jscpd src/ --threshold 5 --format python

  quality-gate:
    needs: [pre-commit, typecheck, test, duplication]
    if: always()
    runs-on: ubuntu-latest
    steps:
      - run: |
          for result in "${{ needs.pre-commit.result }}" "${{ needs.typecheck.result }}" \
                        "${{ needs.test.result }}" "${{ needs.duplication.result }}"; do
            [[ "$result" != "success" ]] && echo "Gate failed" && exit 1
          done

Key principles: pyproject.toml as single source of truth, cache pre-commit and mypy/ruff caches in CI, --output-format=github in Ruff for PR annotations, keep pre-commit under 5 seconds total.

Anti-Patterns

• Skipping cognitive complexity because xenon "already covers it" — they measure different things
• Running all tools in pre-commit — stage slow tools to pre-push or CI-only
• No diff-cover — overall coverage hides untested new code
• No architectural enforcement — dependency violations are the most expensive defects
• Same thresholds for AI-heavy codebases — tighten them (complexity 8, duplication 3%, statements 30)
• Ignoring deptry — transitive dependency reliance is a ticking time bomb
• No mutation testing on critical paths — coverage ≠ test quality

Source: sam-dumont/claude-skills — distributed by TomeVault.