Python Quality Gate Skill

Run four quality tools in deterministic order -- ruff, pytest, mypy, bandit -- and produce a structured pass/fail report with severity-categorized issues and auto-fix commands.

Reference Loading Table

Instructions

Phase 1: Detection and Setup

Step 1: Read CLAUDE.md and detect project configuration.

Read and follow the repository's CLAUDE.md before any execution. Then detect project configuration:

ls -la pyproject.toml setup.py setup.cfg mypy.ini .python-version 2>/dev/null

Identify Python version target, ruff config, pytest config, mypy config from pyproject.toml. Only validate code -- never add tools, features, or flexibility not requested.

Step 2: Detect source and test directories.

ls -d src/ app/ lib/ 2>/dev/null || echo "Source: current directory"
ls -d tests/ test/ 2>/dev/null || echo "Tests: not found"

Step 3: Verify tool availability.

ruff --version
pytest --version
mypy --version || echo "mypy not installed (optional)"
bandit --version || echo "bandit not installed (optional)"

If ruff or pytest are missing, STOP. These are required:

ERROR: Required tool not found: {tool_name}
Install with: pip install ruff pytest pytest-cov

Do not install missing tools automatically unless the user explicitly requests it. Do not modify pyproject.toml or configuration files unless explicitly asked.

Gate: ruff and pytest available. Project structure identified. Proceed only when gate passes.

Phase 2: Execute Quality Checks

Run all checks in fixed order, capturing full output for each. Show complete command output with exact file paths and line numbers -- never summarize or paraphrase tool output, because summarization hides the details engineers need to locate and fix issues.

Step 1: Ruff linting.

ruff check . --output-format=grouped

Step 2: Ruff formatting check.

ruff format --check .

Step 3: Type checking with mypy (if installed).

mypy . --ignore-missing-imports --show-error-codes

Skip and note in report if mypy is not installed. Even if tests pass, still run mypy when available -- tests check behavior while types check contracts, and passing one does not make the other redundant.

Step 4: Run test suite.

pytest -v --tb=short --cov=src --cov-report=term-missing

If no tests directory exists, skip and note in report. Never skip tests to make the gate pass -- tests verify functionality, and skipping them hides broken code. Only skip optional tools (mypy, bandit) if genuinely unavailable, not to manufacture a passing status.

Step 5: Security scanning with bandit (if installed).

bandit -r src/ -ll --format=screen

Skip and note in report if bandit is not installed. Linting passing does not mean code is correct -- linting finds style issues, not logic or security bugs. Run every available tool.

Gate: All available tools have been run. Full output captured. Proceed to analysis.

Phase 3: Categorize and Analyze

Step 1: Categorize issues by severity.

See references/tool-commands.md for complete severity classification tables.

Summary of severity levels:

• Critical: F errors (pyflakes), E9xx (syntax), undefined names, test failures, high-severity security
• High: E501, E711/E712, F841, N8xx, arg-type/assignment mypy errors
• Medium: W warnings, C4xx, no-untyped-def mypy errors
• Low: SIM suggestions, UP upgrade suggestions

Always prioritize critical issues over style fixes -- critical issues (F errors, test failures) break functionality while style issues do not. Fix critical first, high second; use auto-fix for bulk style cleanup only after critical issues are resolved.

Step 2: Count auto-fixable issues.

ruff check . --statistics

Issues marked with [*] are auto-fixable. Show suggested auto-fix commands for these issues so users know what can be fixed automatically.

Step 3: Determine overall status.

FAIL if:

• Any ruff F errors or test failures
• Any high-severity bandit issues
• Mypy errors exceed 10 (configurable)
• Test coverage below 80% (if coverage enabled)

PASS otherwise. Exit with non-zero status if any critical check fails.

Gate: All issues categorized. Pass/fail determined. Proceed to report.

Phase 4: Generate Report

Format a structured markdown report. See references/report-template.md for the full template.

The report MUST include:

1. Overall PASS/FAIL status
2. Summary table (each tool's status and issue count)
3. Total issues and auto-fixable count
4. Detailed results per tool (issues grouped by severity, then grouped by type and file for readability)
5. Critical issues requiring attention with file:line references
6. Auto-fix commands section
7. Quality metrics: error counts and coverage percentages

Report facts -- show raw command output rather than describing it. No self-congratulation ("great job", "looking good"). Generate the full report even when only style issues are found, because style issues can hide real problems in noise and a full severity-prioritized report surfaces them.

Print the complete report to stdout. Never summarize or truncate. If --output {file} flag was provided, also write report to file. Remove any intermediate temporary files at completion -- keep the final report only if the user requested file output.

Gate: Report generated and displayed. Task complete.

Auto-Fix Mode (only when explicitly requested)

Auto-fix modifies files in place -- never run it without explicit user confirmation. Running ruff --fix blindly can change code semantics (import removal, reformatting), so always run check-only first, review issues, confirm auto-fix intent, then verify changes.

When user explicitly requests auto-fix:

ruff check . --fix
ruff format .

After auto-fix, show the diff so changes can be reviewed, then re-run the quality gate to verify:

git diff

Examples

Example 1: Pre-Merge Quality Check

User says: "Run quality checks before I merge this PR" Actions:

1. Detect project config and available tools (Phase 1)
2. Run ruff check, ruff format, mypy, pytest, bandit in order (Phase 2)
3. Categorize 12 issues: 0 critical, 3 high, 9 medium (Phase 3)
4. Generate report showing PASSED with 3 high-priority suggestions (Phase 4)

Result: Structured report with actionable fix commands

Example 2: Quality Gate Failure

User says: "Check code quality on the payments module" Actions:

1. Detect config, find src/payments/ directory (Phase 1)
2. Run all tools -- pytest shows 2 failures, ruff finds F401 errors (Phase 2)
3. Categorize: 2 critical (test failures), 1 critical (undefined name), 5 medium (Phase 3)
4. Generate FAILED report with critical issues listed first (Phase 4)

Result: FAILED status with prioritized fix list, auto-fix commands for 5 medium issues

Error Handling

Error: "ruff: command not found"

Cause: Ruff is not installed in the current environment Solution: Install with pip install ruff. Do not proceed without ruff -- exit with status 2.

Error: "Tests failed with exit code 1"

Cause: pytest found test failures Solution: This is expected behavior, not a tool error. Parse output, include failure details in report, mark overall status as FAILED, continue with remaining checks.

Error: "No Python files found"

Cause: Running from wrong directory or not a Python project Solution: Verify location with ls pyproject.toml src/ tests/. Run from project root.

Error: "Mypy cache corruption"

Cause: Stale or corrupted .mypy_cache directory Solution: Clear cache with rm -rf .mypy_cache and retry. If mypy continues to fail, skip type checking and note in report.

References

Reference Files

• ${CLAUDE_SKILL_DIR}/references/tool-commands.md: Severity classifications, expected output formats, CLI flags
• ${CLAUDE_SKILL_DIR}/references/report-template.md: Full structured report template
• ${CLAUDE_SKILL_DIR}/references/pyproject-template.toml: Complete ruff, pytest, mypy, bandit configuration