quality-check（质量检查）

Code Quality Check Skill

⚠️ MANDATORY SKILL - AUTO-INVOKED BY SAFE-COMMIT

Purpose

Enforce code quality standards through automated linting, formatting, and static analysis. Ensures code meets project conventions before committing.

CRITICAL: This skill is automatically invoked by safe-commit. NEVER run linters manually.

🚫 COMMIT BLOCKING RULE - ZERO TOLERANCE

Commits are FORBIDDEN unless the LATEST quality check run shows ZERO issues.

This means:

1. If quality check finds errors → FIX THEM
2. After fixing → RERUN quality check (MANDATORY)
3. If rerun finds errors → FIX AND RERUN AGAIN
4. Repeat until clean pass
5. Only after a clean pass may you proceed to commit

The last quality check output MUST show all checks passing. No exceptions.

CRITICAL: Full Repository Scope

All quality checks run on the ENTIRE repository:

• ruff check. - ALL Python files
• eslint. - ALL JS/TS files
• golangci-lint run./... - ALL Go files

You MUST fix ALL issues found, not just those in files you modified. Whoever runs quality-check is responsible for fixing discovered issues.

🚫 NEVER DO THIS

• ❌ Running eslint or npm run lint manually before commit
• ❌ Running golangci-lint run manually before commit
• ❌ Running flake8 or black manually before commit
• ❌ Running linters outside of this skill during commit workflow

Let safe-commit invoke this skill automatically. Manual linting before commit is REDUNDANT and FORBIDDEN.

⚠️ SKILL GUARD - READ BEFORE RUNNING LINTERS MANUALLY

Before using Bash tool to run linters, answer these questions:

❓ Are you about to run npm run lint or eslint?

→ STOP. Are you doing this before commit? If YES, use safe-commit instead (it invokes this skill).

❓ Are you about to run golangci-lint run?

→ STOP. Are you doing this before commit? If YES, use safe-commit instead (it invokes this skill).

❓ Are you about to run flake8, black, or mypy?

→ STOP. Are you doing this before commit? If YES, use safe-commit instead (it invokes this skill).

❓ Are you about to run prettier --check or formatting tools?

→ STOP. Are you doing this before commit? If YES, use safe-commit instead (it invokes this skill).

❓ Are you checking code quality before committing?

→ STOP. Invoke safe-commit skill (it will invoke this skill automatically).

IF YOU RUN LINTERS MANUALLY BEFORE COMMIT, YOU ARE CREATING REDUNDANCY AND WASTING TIME.

When to run linters manually:

• ✅ During development/refinement (not for commit)
• ✅ When user explicitly requests quality check

When NOT to run linters manually:

• ❌ Before commit (use safe-commit instead)
• ❌ As part of commit workflow (use safe-commit instead)

Safe-commit invokes this skill automatically. Don't duplicate the work.

Philosophy

CRITICAL: Treat linter issues as syntax errors, not warnings.

• All linter issues MUST be fixed before commit
• No exceptions unless explicitly approved by user
• Auto-fix when tools support it
• Manual fix when auto-fix unavailable

⚠️ CRITICAL: Rerun Policy - ZERO TOLERANCE (All Languages)

Assuming fixes resolved all errors is FORBIDDEN.

This applies to ALL auto-fix operations in ANY language:

• ESLint --fix, Prettier --write (Node.js/TypeScript)
• gofmt -w, golangci-lint --fix (Go)
• ruff --fix, ruff format, black, isort (Python)
• cargo fmt, cargo clippy --fix (Rust)
• spotless:apply (Java)

After ANY Fix (Auto or Manual):

1. MUST rerun the same check to verify fix succeeded
2. MUST NOT proceed until rerun passes
3. MUST NOT assume any fix caught everything
4. MUST repeat fix → rerun cycle until clean

The Fix-Rerun Loop (MANDATORY)

while (quality_check_has_errors) {
    fix_errors();           // Auto-fix or manual
    rerun_quality_check();  // ALWAYS rerun after fixing
}
// Only exit loop when rerun shows ZERO errors

Why This Matters

• Auto-fix tools don't catch 100% of issues
• Manual fixes can introduce new issues
• Some fixes reveal previously hidden issues
• Proceeding without verification can commit broken code
• Commits with quality issues pollute the codebase

Example (WRONG) - Applies to ALL Languages

❌ Found 5 linting errors
❌ Running auto-fix...
❌ "Fixed! Proceeding to commit..."  ← FORBIDDEN - DID NOT RERUN

Example (ALSO WRONG) - Manual Fix Without Rerun

❌ Found 3 type errors
❌ [Manually fixed the code]
❌ "Fixed! Proceeding to commit..."  ← FORBIDDEN - DID NOT RERUN

Example (CORRECT) - Full Fix-Rerun Loop

✅ Found 5 linting errors
✅ Running auto-fix...
✅ Re-running lint check...          ← MANDATORY RERUN
✅ Found 1 remaining error (requires manual fix)
✅ [Fix manually]
✅ Re-running lint check...          ← MANDATORY RERUN AGAIN
✅ All checks pass. Safe to proceed.

Workflow (Quick Summary)

Core Steps

1. Detect Languages: Check for package.json, go.mod, requirements.txt, Cargo.toml, etc.
2. Run Standard Checks: Linting, formatting, type checking, static analysis (auto-fix when possible)
3. Run Deep Audits: Language-specific skills (control-flow-check, type-safety-audit, n-plus-one-detection)
4. Handle Results: Report pass/auto-fix/fail, provide specific error locations and fixes
5. Verify Multi-Language: All languages must pass before commit

Language Support

• Node.js/TypeScript: ESLint, Prettier, tsc, npm audit
• Go: gofmt, go vet, golangci-lint, go mod tidy
• Python: ruff (lint+format), mypy | fallback: black, flake8, isort
• Rust: rustfmt, clippy
• Java: Maven/Gradle spotless, checkstyle

For detailed language-specific commands and checks:

Read `~/.claude/skills/quality-check/references/LANGUAGE-CHECKS.md`

Use when: Need specific tool commands, troubleshooting linter issues, or setting up new language

For deep audit integration (control-flow, type-safety, N+1 detection):

Read `~/.claude/skills/quality-check/references/DEEP-AUDITS.md`

Use when: Running language-specific deep audits or integrating specialized checks

For result handling and reporting patterns:

Read `~/.claude/skills/quality-check/references/RESULT-HANDLING.md`

Use when: Formatting output, handling auto-fix scenarios, or reporting manual fix requirements

Integration with Other Skills

This skill is invoked by:

• safe-commit - Before committing changes
• create-pr - Before creating pull requests

Best Practices

1. Run in parallel - Multiple languages? Check simultaneously
2. Auto-fix first - Always attempt auto-fix before asking user
3. Be specific - Show exact file, line, and error
4. Suggest solutions - Don't just report errors, help fix them
5. Verify fixes - Re-run checks after auto-fixing
6. No skip option - Quality checks are mandatory

Tool Installation Detection

Before running tools, check if they're available:

# Example for Node.js:
if ! command -v eslint &> /dev/null; then
    echo "ESLint not found. Install with: npm install -D eslint"
    exit 1
fi

If tool missing:

• Report missing tool
• Provide installation command
• Ask user to install or use project's package.json scripts

Configuration Detection

Check for project-specific configuration:

• .eslintrc.js, .eslintrc.json (ESLint)
• .prettierrc, prettier.config.js (Prettier)
• .golangci.yml (golangci-lint)
• pyproject.toml (Python tools)
• .editorconfig (Cross-language)

Use project config when available, otherwise use sensible defaults.

Handling Make/NPM Scripts

If project has quality scripts in Makefile or package.json:

# Check for make target:
make lint
make fmt

# Check for npm script:
npm run lint
npm run format

Prefer project scripts over direct tool invocation when available.

Emergency Override

Only if user explicitly states "skip quality checks" or "I acknowledge quality issues":

1. Document the override in output
2. List specific issues being ignored
3. Suggest creating follow-up ticket
4. Proceed with commit

This should be RARE - quality checks exist for good reason.