Linter-Driven Development Workflow (TypeScript + React)

DEFAULT ENTRY POINT: This skill should be automatically invoked for ALL code changes when the plugin is enabled.

META ORCHESTRATOR for implementation workflow: design → test → lint → refactor → review → commit. Use for any commit: features, bug fixes, refactors.

Auto-Invocation

When this plugin is enabled in a project, this skill is the default workflow for:

New feature implementation
Bug fixes
Refactoring
Any code change that should result in a commit

The CLAUDE.md file at the plugin root enforces this automatic invocation.

Core Principles

CRITICAL - Read Before Starting:

Never Disable Rules by Default
- DO NOT add eslint-disable, @ts-ignore, @ts-expect-error, or similar comments without explicit user approval
- ALWAYS strive to fix the underlying issue through refactoring, not by suppressing warnings
- ONLY disable rules as an absolute last resort when no other solution exists, and always ask user first
- When approved: Add a comment explaining WHY the rule is disabled
Fix, Don't Suppress
- Linter warnings exist for good reasons - they indicate real issues
- Use @refactoring patterns to address complexity and code quality issues
- If a rule seems wrong for the project, discuss with user rather than silently disabling
Clean Code at Every Step
- All checks must pass without suppressions before commit
- Tests pass, linter passes, types pass - no exceptions

Prerequisites

IMPORTANT: Before using this skill, the project MUST have linter configurations:

Required Configurations

TypeScript (tsconfig.json)
- Type checking configured for your project
ESLint (eslint.config.mjs or .eslintrc.js)
- Must include eslint-plugin-sonarjs for complexity metrics
- Recommended: TypeScript ESLint, React plugins
- Complexity thresholds: cognitive, cyclomatic, expression
Prettier (.prettierrc.json or prettier.config.js or .prettierrc)
- Consistent formatting rules defined
- Integration with ESLint recommended
Stylelint (stylelint.config.js) - if using CSS/SCSS
- CSS/SCSS linting rules configured

Required npm Scripts

Project must have scripts for running quality checks. Script names vary by project - detect them from package.json.

Common patterns to look for:

Testing: test, test:unit, vitest, jest
Type checking: typecheck, type-check, tsc, check-types
Linting (check): lint, lint:check, eslint, lintcheck
Linting (fix): lint:fix, eslint:fix, lint --fix
Formatting (check): format, format:check, prettier:check, formatcheck
Formatting (fix): format:fix, prettier:write, prettier --write
Styling (check): stylelint, style:check, stylecheck
Styling (fix): stylelint:fix, style:fix
Combined check: check, checkall, validate, verify
Combined fix: fix, fixall, format:all

Detection strategy: Read package.json scripts and identify which commands serve each purpose.

Step 0: Detect Project Setup

FIRST STEP - Do this before starting any workflow:

Detect Package Manager
- Check for lock files in project root:
  - yarn.lock → Use yarn commands
  - package-lock.json or npm-shrinkwrap.json → Use npm commands
  - pnpm-lock.yaml → Use pnpm commands
  - No lock file → Ask user which package manager to use
Detect Available Scripts
- Read package.json scripts section
- Identify which scripts exist for each purpose:
  - Type checking (e.g., typecheck, type-check, tsc)
  - Linting check (e.g., lint, eslint, lint:check)
  - Linting fix (e.g., lint:fix, eslint:fix)
  - Formatting check (e.g., format:check, prettier:check)
  - Formatting fix (e.g., format:fix, prettier:write)
  - Testing (e.g., test, vitest, jest)
  - Combined checks (e.g., check, checkall, validate)
  - Combined fixes (e.g., fix, fixall)
Build Command Map
- Store detected commands for use throughout workflow
- Example: { typecheck: 'typecheck', lint: 'lint', lintFix: 'lint:fix', test: 'test' }

Remember: Use detected package manager + detected script names consistently throughout ALL workflow phases.

Verification

Before starting, verify setup by running detected commands:

Type checking works (using detected typecheck script)
Linting works (using detected lint script)
Formatting works (using detected format script)
Tests work (using detected test script)
SonarJS plugin installed and configured

If scripts are missing:

Check if functionality exists but with different script name
Look for combined commands (e.g., check that runs multiple tools)
If truly missing, ask user:
- "What command should I run to check/fix linting?"
- "Where is this documented?" (suggest adding to README.md or CLAUDE.md)
If no script exists, run tool directly (e.g., eslint .) or skip that phase

Command detection examples:

# If package.json has:
"scripts": {
  "lint": "eslint .",
  "lint:fix": "eslint . --fix",
  "check": "tsc && eslint .",
  "test": "vitest"
}

# Detected commands:
typecheck: (not found, will run 'tsc' directly)
lint: 'lint'
lintFix: 'lint:fix'
test: 'test'
combined: 'check'

When to Use

Implementing any code change that should result in a commit
Need automatic workflow management with quality gates
Want to ensure: clean code + tests + linting + design validation + accessibility

Workflow Phases

IMPORTANT: Start every workflow by detecting the package manager (Step 0 in Prerequisites).

Phase 1: Design (if needed)

If new components/types/major changes needed → invoke @component-designing skill
Output: Component design plan with types, hooks, and structure

Phase 2: Implementation

Follow @testing skill principles (React Testing Library with Jest/Vitest)
Write tests + implementation in parallel (not necessarily test-first)
Follow project's Prettier/ESLint formatting rules
Use project's test runner (Jest, Vitest, or other)
Aim for 100% coverage on new leaf components/hooks (pure logic with no external dependencies)
- Leaf types: Pure logic (can compose other leaf types), no API/DB/file system access
- Orchestrating types: Coordinate leaf types and external systems, need integration tests
Test from user perspective (public API only)

Phase 3: Linter Loop

Use detected package manager and script names from Step 0 for all commands below.

Run quality checks in this order (using detected script names from package.json):

Type Check: Run detected typecheck script (look for: typecheck, type-check, tsc)
Lint Check: Run detected lint check script (look for: lint, lint:check, eslint)
Format Check: Run detected format check script (look for: format:check, prettier:check)
Style Check: Run detected style check script (look for: stylelint, style:check) - if CSS/SCSS in project

Handling missing scripts:

If type check script not found → Run tsc --noEmit directly
If lint check script not found → Run eslint . directly
If format check script not found → Run prettier --check . directly
If style check script not found and CSS/SCSS exists → Skip if Stylelint not installed

If any failures detected:

Run auto-fixes using detected fix scripts:
- Lint fix: Run detected lint fix script (look for: lint:fix, eslint:fix)
- Format fix: Run detected format fix script (look for: format:fix, prettier:write)
- Style fix: Run detected style fix script (look for: stylelint:fix, style:fix)
Re-run quality checks
If still failing (complexity, design issues):
- NEVER disable linter rules by default: Do NOT add eslint-disable, @ts-ignore, @ts-expect-error, or similar comments unless explicitly approved by the user
- Always fix, not disable: Strive to fix the underlying issue through refactoring, not by suppressing warnings
- Only disable as last resort: If absolutely necessary and no other solution exists, ask user for explicit approval before disabling any rule
- When approved: Add a comment explaining WHY the rule is disabled
- Interpret failures (cognitive complexity, cyclomatic complexity, etc.)
- Invoke @refactoring skill to fix (use storifying, extract functions/hooks, early returns)
- Check for existing utilities: Before creating new helpers, check if type guards, utilities, or constants already exist in the codebase
- Avoid repeated typeof checks: Use or create type guard utilities (e.g., isString, isNumber) instead of repeating typeof value === 'string'
- Re-run checks
Repeat until all checks pass clean

Alternative: If project has combined commands (detected in Step 0):

Check: Use detected combined check script
Fix: Use detected combined fix script

Example workflow:

# Step 0: Detect commands from package.json
# Found: { packageManager: 'npm', typecheck: 'typecheck', lint: 'lint', lintFix: 'lint:fix', ... }

# Run checks using detected script names
[package-manager] run [detected-typecheck-script]
[package-manager] run [detected-lint-script]
[package-manager] run [detected-format-check-script]

# If failures, run fixes using detected fix scripts
[package-manager] run [detected-lint-fix-script]
[package-manager] run [detected-format-fix-script]

# Re-run checks to verify
[package-manager] run [detected-typecheck-script]
[package-manager] run [detected-lint-script]
[package-manager] run [detected-format-check-script]

Phase 4: Pre-Commit Design Review (ADVISORY)

Invoke @pre-commit-review skill
Review validates design principles (not code correctness)
Includes accessibility checks (ARIA, semantic HTML, keyboard nav)
Categorized findings: Design Debt / Readability Debt / Polish Opportunities
If issues found in broader file context, flag for potential refactor
User decides: commit as-is, apply fixes, or expand scope

Phase 5: Commit Ready

Type checking passes ✅
ESLint passes ✅
Prettier passes ✅
Stylelint passes ✅
Tests pass with target coverage ✅
Design review complete (advisory) ✅
Present summary + commit message suggestion

Output Format

📋 COMMIT READINESS SUMMARY

✅ Type Check: Passed (0 errors)
✅ ESLint: Passed (0 issues)
✅ Prettier: Passed (all files formatted)
✅ Stylelint: Passed (0 style issues)
✅ Tests: 92% coverage (3 leaf hooks at 100%, 1 orchestrating component, 18 test cases)
⚠️  Design Review: 3 findings (see below)

🎯 COMMIT SCOPE
Modified:
- src/components/LoginForm.tsx (+65, -20 lines)
- src/hooks/useAuth.ts (+30, -5 lines)

Added:
- src/types/auth.ts (new: UserId, Email types)
- src/contexts/AuthContext.tsx (new context provider)

Tests:
- src/components/LoginForm.test.tsx (+95 lines)
- src/hooks/useAuth.test.ts (new)
- src/types/auth.test.ts (new)

⚠️  DESIGN REVIEW FINDINGS

🔴 DESIGN DEBT (Recommended to fix):
- src/components/LoginForm.tsx:45 - Primitive obsession detected
  Current: function validateEmail(email: string): boolean
  Better:  Use Zod schema or branded Email type with validation
  Why: Type safety, validation guarantee, prevents invalid emails
  Fix: Use @component-designing to create self-validating Email type

- src/hooks/useAuth.ts:78 - Prop drilling detected
  Auth state passed through 3+ component levels
  Why: Tight coupling, hard to maintain
  Fix: Extract AuthContext or use composition pattern

🟡 READABILITY DEBT (Consider fixing):
- src/components/LoginForm.tsx:120 - Mixed abstraction levels
  Component mixes validation logic with UI rendering
  Why: Harder to understand and test independently
  Fix: Use @refactoring to extract custom hooks (useValidation)

- src/components/LoginForm.tsx:88 - Cognitive complexity: 18 (max: 15)
  Nested conditionals for form validation
  Why: Hard to understand logic flow
  Fix: Use @refactoring to extract validation functions or use Zod

🟢 POLISH OPPORTUNITIES:
- src/types/auth.ts:12 - Missing JSDoc comments
  Public types should have documentation
- src/components/LoginForm.tsx:45 - Consider semantic HTML
  Use <form> with proper ARIA labels for better accessibility
- src/hooks/useAuth.ts:34 - Missing error boundaries
  Consider wrapping async operations with error handling

📝 BROADER CONTEXT:
While reviewing LoginForm.tsx, noticed similar validation patterns in
RegisterForm.tsx and ProfileForm.tsx (src/components/). Consider
extracting a shared validation hook or creating branded types for common
fields (Email, Username, Password) used across the application.

💡 SUGGESTED COMMIT MESSAGE
Add self-validating Email and UserId types to auth feature

- Introduce Email type with RFC 5322 validation using Zod
- Introduce UserId branded type for type safety
- Refactor LoginForm to use validated types
- Extract useAuth hook for auth state management
- Add AuthContext to eliminate prop drilling
- Achieve 92% test coverage with React Testing Library

Follows component composition principles and reduces primitive obsession.

────────────────────────────────────────

Would you like to:
1. Commit as-is (ignore design findings)
2. Fix design debt only (🔴), then commit
3. Fix design + readability debt (🔴 + 🟡), then commit
4. Fix all findings (🔴 🟡 🟢), then commit
5. Refactor broader scope (address validation patterns across features), then commit

Complexity Thresholds (SonarJS)

These metrics trigger @refactoring when exceeded:

Cognitive Complexity: max 15 (how hard to understand)
Cyclomatic Complexity: max 10 (number of paths through code)
Expression Complexity: max 5 (operators in single expression)
Function Length: max 200 lines
File Length: max 600 lines
Nesting Level: max 4 (depth of nested control structures)
Max Union Size: max 4 types (union types with too many options)

Priority Levels

🔴 High Priority: Type Safety Issues

Rules that can cause runtime errors:

@typescript-eslint/no-unsafe-member-access (416 violations)
@typescript-eslint/no-unsafe-assignment (267 violations)
@typescript-eslint/no-explicit-any (194 violations)
@typescript-eslint/no-unsafe-argument (101 violations)
@typescript-eslint/no-unsafe-call (72 violations)
@typescript-eslint/no-unsafe-return (50 violations)

Fix Strategy: Use proper types, type guards, Zod schemas, or branded types

🟡 Medium Priority: Code Quality & Maintainability

Rules that affect readability and maintenance:

no-magic-numbers (243 violations) - Extract to named constants
react/forbid-dom-props (94 violations) - No inline styles, use CSS modules
sonarjs/cyclomatic-complexity (34 violations) - Reduce branches, early returns
sonarjs/prefer-read-only-props (25 violations) - Props should be immutable
react-hooks/exhaustive-deps (41 violations) - Fix dependencies or simplify

Fix Strategy: Apply @refactoring patterns (storifying, early returns, extract functions)

🟢 Low Priority: Style & Convention

Rules that improve consistency:

no-console (28 violations) - Use proper logging
Import/export conventions
Styling conventions (camelCase, keyframes naming)

Fix Strategy: Auto-fix or manual cleanup

Workflow Control

Sequential Phases: Each phase depends on previous phase completion

Design must complete before implementation
Implementation must complete before linting
Linting must pass before review
Review must complete before commit

Iterative Linting: Phase 3 loops until clean Advisory Review: Phase 4 never blocks, always asks user

Integration with Other Skills

This orchestrator invokes other skills automatically:

@component-designing (Phase 1, if needed)
@testing (Phase 2, principles applied)
@refactoring (Phase 3, when linter fails on complexity)
@pre-commit-review (Phase 4, always)

After committing, consider:

If feature complete → invoke @documentation skill
If more work needed → run this workflow again for next commit

Common Linter Failures and Resolutions

TypeScript Errors (detected typecheck script)

Type mismatches → Fix types or add proper type guards
Missing types → Add explicit types or interfaces
Cannot fix automatically → Manual intervention required

ESLint Failures (detected lint check script)

Auto-fixable:

Import sorting (simple-import-sort)
Unused imports (unused-imports)
Formatting issues covered by Prettier
Simple style violations

Requires refactoring (invoke @refactoring):

Cognitive/cyclomatic complexity
Max lines per function
Expression complexity
Nested control flow
React hooks violations
Component design issues

Prettier Failures (detected format check script)

Always auto-fixable with detected format fix script or prettier --write .
No manual intervention needed

Stylelint Failures (detected style check script)

Most auto-fixable with detected style fix script or stylelint "**/*.{css,scss}" --fix
Class naming violations may require manual fixes

Best Practices

Run checks frequently during development
Fix one complexity issue at a time (don't batch refactoring)
Trust the advisory review (design debt causes future pain)
Test after each refactoring (ensure behavior unchanged)
Commit frequently (small, focused commits)

Implementation Phases

Phase 1: Type Safety Foundation (🔴 High Priority)

Fix all @typescript-eslint/no-unsafe-* errors
Replace any types with proper types
Add type definitions for external APIs
Use type guards for runtime checks
Implement Zod schemas for validation

Phase 2: Code Quality (🟡 Medium Priority)

Fix complexity issues (cyclomatic, cognitive)
Remove inline styles, use CSS modules
Make props readonly
Extract magic numbers to constants
Apply refactoring patterns (IIFE → lookup, empty blocks → early returns)

Phase 3: Polish (🟢 Low Priority)

Acceptance Criteria

CRITICAL: All criteria must be met before completing this skill.

Mandatory Requirements (Must Pass)

No Linter Rule Disabling in Changed Files
- Changed files contain NO new eslint-disable, eslint-disable-next-line, eslint-disable-line comments
- Changed files contain NO new @ts-ignore, @ts-expect-error, @ts-nocheck comments
- Changed files contain NO new stylelint-disable comments
- Pre-existing disabling comments in unchanged files are acceptable (not in scope)
- If disabling is absolutely necessary: Stop and ask user for explicit approval before adding
- When approved: Add a comment explaining WHY the rule is disabled
All Quality Checks Pass Clean
- TypeScript compilation: 0 errors
- ESLint: 0 errors, 0 warnings in changed files
- Prettier: All changed files formatted correctly
- Stylelint: 0 errors in changed CSS/SCSS files (if applicable)
All Tests Pass
- Existing tests pass (no regressions)
- New tests written for new code
- Coverage targets met (100% for leaf components/hooks)
Iterative Verification
- Run quality checks at least twice after final changes
- Second run confirms no new issues introduced
- If any check fails on second run, fix and repeat until two consecutive clean runs

Verification Workflow

IMPORTANT: Detect available scripts from the project's package.json before running checks.

# Iteration 1: Initial check
Run all quality check commands detected from package.json:
- TypeScript check (e.g., typecheck, type-check, tsc)
- Linting check (e.g., lint, lint:check, eslint)
- Format check (e.g., format:check, prettier:check)
- Tests (e.g., test, test:unit, vitest)

# If failures: fix issues, then...

# Iteration 2: Verify fixes didn't introduce new issues
Run the same quality check commands again

# If still clean: proceed to commit
# If new failures: fix and repeat until two consecutive clean runs

Pre-Commit Checklist

Before marking workflow complete:

✅ ACCEPTANCE CRITERIA CHECKLIST

Linter Compliance:
[ ] No eslint-disable comments added to changed files
[ ] No @ts-ignore/@ts-expect-error added to changed files
[ ] All linter issues fixed through proper refactoring
[ ] If any disabling approved: comment explains WHY

Quality Gates:
[ ] TypeScript: 0 errors
[ ] ESLint: 0 errors/warnings in changed files
[ ] Prettier: All files formatted
[ ] Tests: All passing

Iteration Verification:
[ ] Ran checks twice consecutively
[ ] Both runs passed clean
[ ] No oscillating fixes (fix A breaks B, fix B breaks A)

Ready to commit: All boxes checked ✅

What Blocks Completion

The following will BLOCK skill completion:

Any new linter disabling comment in changed files (without explicit user approval)
Any failing quality check (typecheck, lint, format, test)
Single-run verification (must run checks twice)
Unresolved complexity issues (must refactor, not disable)

Acceptable Exceptions (Require User Approval)

Only with explicit user consent:

Disabling a rule for a specific line with documented justification
Skipping a quality check due to known project issues
Accepting technical debt with plan to address later

When disabling a rule with approval, add a comment explaining WHY:

// ❌ Bad: Disabled without explanation
// eslint-disable-next-line @typescript-eslint/no-explicit-any
const data: any = response.body

// ✅ Good: Disabled with justification
// eslint-disable-next-line @typescript-eslint/no-explicit-any -- Legacy API returns untyped response, migration planned in JIRA-123
const data: any = response.body

Document any exceptions in commit message.

Additional Resources

Common Linter Failures section above for resolution strategies
@refactoring skill - For complexity issues (cognitive, cyclomatic, expression)
@component-designing skill - For type safety and architecture issues
@pre-commit-review skill - For design validation (runs automatically in Phase 4)

linter-driven-development