shared-tooling-git-hooks

Git Hooks

Quick Guide: Husky v9 for git hooks with "prepare": "husky" (NOT "husky install"). lint-staged v16 for staged-only linting. commitlint for conventional commit messages. Pre-commit hooks should take < 10 seconds. Set HUSKY=0 in CI/production.

---

<critical_requirements>

CRITICAL: Before Using This Skill

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST use "prepare": "husky" in package.json - NOT the deprecated "husky install")

(You MUST only lint staged files via lint-staged - NEVER lint the entire codebase in pre-commit)

(You MUST set HUSKY=0 in CI/production environments to disable hooks)

(You MUST use plain hook files in .husky/ directory - NO shebang lines or husky.sh sourcing in v9)

(You MUST keep pre-commit hooks under 10 seconds - move slow tasks to pre-push or CI)

</critical_requirements>

---

Auto-detection: Husky, husky init, .husky/, pre-commit hook, lint-staged, commitlint, conventional commits, commit-msg hook, git hooks, prepare script

When to use:

• Setting up pre-commit hooks with Husky + lint-staged
• Configuring commit message validation with commitlint
• Migrating from Husky v8 to v9
• Configuring git hooks in monorepo setups
• Disabling hooks in CI/production environments

When NOT to use:

• Linter/formatter configuration itself (separate concern)
• CI/CD pipeline configuration (separate concern)
• Runtime application code (this is developer workflow tooling only)

Key patterns covered:

• Husky v9 setup and hook creation
• lint-staged v16 configuration patterns
• commitlint with conventional commits
• CI/production hook disabling
• Monorepo setup
• Migration from Husky v8 to v9

Detailed Resources:

• examples/core.md - Setup, lint-staged config, commitlint, CI handling, monorepo, migration
• reference.md - Decision frameworks, tool comparison, anti-patterns

---

Philosophy

Git hooks are a developer workflow tool - they catch issues early while staying fast and non-blocking. The goal is fast feedback (< 10 seconds) on staged files only. Hooks are optional infrastructure; many projects work fine without them.

When to use git hooks:

• Team projects where code quality gates prevent CI failures
• Projects with established linting/formatting that should be enforced
• When you want fast feedback before code reaches CI
• Monorepos where running full lint is too slow

When NOT to use:

• Solo projects where you always remember to lint (overhead without benefit)
• Projects without established linting/formatting rules yet (set up linting first)
• When pre-commit hooks exceed 10 seconds (move to CI instead)
• CI-only projects where hooks add friction without value

---

Core Patterns

Pattern 1: Husky v9 Setup

Husky v9 uses plain shell scripts in .husky/ directory. No shebang lines needed. The prepare script auto-installs hooks for all team members.

# Full setup in 4 commands
bun add -D husky
bunx husky init       # Creates .husky/ and adds "prepare": "husky" to package.json
bun add -D lint-staged
echo "bunx lint-staged" > .husky/pre-commit

Key points:

• "prepare": "husky" (NOT "husky install" - deprecated, will break in v10)
• Hook files are plain shell scripts (no shebang required in v9)
• HUSKY=0 disables hooks (CI/production); HUSKY=2 enables debug mode
• v9.1.1+ allows running package commands directly without npx/bunx

See examples/core.md for full setup and package.json configuration.

---

Pattern 2: Pre-commit Hook with lint-staged

lint-staged v16 runs commands only on staged files. Uses picomatch for glob matching (replaced micromatch).

// lint-staged.config.mjs
export default {
  "*.{ts,tsx}": ["eslint --fix", "prettier --write"],
  "*.{css,scss}": ["prettier --write"],
};

Type checking requires function syntax (runs on ALL files, not just staged):

// lint-staged.config.mjs — with type checking
export default {
  "*.{ts,tsx}": ["eslint --fix", "prettier --write"],
  "*.{ts,tsx,js,jsx}": () => "tsc --noEmit",
};

Why good: Only staged files, auto-fix reduces manual work, fast feedback

v16 breaking changes: --shell flag removed (use shell scripts instead), requires Node.js 20.18+

See examples/core.md for multiple file type patterns and v16 migration details.

---

Pattern 3: Commitlint with Conventional Commits

commitlint v20+ validates commit messages. ESM-native - use .mjs config extension.

bun add -D @commitlint/cli @commitlint/config-conventional

# .husky/commit-msg
bunx commitlint --edit $1

// commitlint.config.mjs (MUST be .mjs for Node v24 compatibility)
export default {
  extends: ["@commitlint/config-conventional"],
};

Format: type(scope): description - types: feat, fix, docs, refactor, test, chore, etc.

See examples/core.md for v20 changes and advanced configuration.

---

Pattern 4: CI/Production Environment Handling

Disable Husky where hooks should not run.

# CI pipelines
HUSKY=0 npm install

# GitHub Actions
env:
  HUSKY: 0

Why: Prevents hook installation failures when devDependencies not installed in production.

See examples/core.md for conditional prepare script alternatives.

---

Pattern 5: Monorepo Setup

For monorepos where package.json is not at the repository root.

// apps/frontend/package.json
{
  "scripts": {
    "prepare": "cd ../.. && husky apps/frontend/.husky"
  }
}

Key: Navigate to repo root, then pass the .husky/ directory path to Husky.

---

Pattern 6: Migration from Husky v8 to v9

Step-by-step: update prepare script, remove shebangs/husky.sh sourcing from hook files, delete .husky/.gitignore, remove pinst if used.

Important: Hooks with deprecated shebang #!/usr/bin/env sh and husky.sh sourcing will fail in v10.0.0. Migrate now.

See examples/core.md for the full migration walkthrough.

---

<decision_framework>

Decision Framework

What to Run Pre-commit

What to run pre-commit?
├─ Fast (< 10 seconds)?
│   ├─ Lint with auto-fix → YES (lint-staged)
│   ├─ Format → YES (lint-staged)
│   └─ Type check (--noEmit) → YES (lint-staged)
└─ Slow (> 10 seconds)?
    ├─ Full test suite → NO (pre-push or CI)
    ├─ Full build → NO (CI)
    └─ E2E tests → NO (CI)

Which Hook for Which Task

Which Git hook?
├─ Before committing? → pre-commit (lint-staged)
├─ Validating commit message? → commit-msg (commitlint)
├─ Before pushing? → pre-push (type check, unit tests)
├─ Before merging? → pre-merge-commit (v9.1.5+)
└─ After checkout/merge? → post-checkout / post-merge (install deps)

Pre-commit Timing Guidelines

Task	Time	Pre-commit?
lint-staged (staged files only)	< 5s	Yes
Code formatting	< 2s	Yes
Type check (--noEmit)	< 10s	Yes
Full test suite	> 30s	No (CI)
E2E tests	> 60s	No (CI)
Full build	> 30s	No (CI)

See reference.md for Husky vs alternatives comparison.

</decision_framework>

---

<red_flags>

RED FLAGS

High Priority Issues:

• Using deprecated "prepare": "husky install" instead of "prepare": "husky" (will break in v10)
• Running full lint on entire codebase in pre-commit hook (too slow, defeats staged-only purpose)
• Using v4-style "husky": { "hooks": {} } config in package.json (not supported in v9)
• Using HUSKY_GIT_PARAMS environment variable (deprecated, use $1 instead)

Medium Priority Issues:

• Pre-commit hooks taking > 10 seconds (encourages --no-verify abuse)
• Missing HUSKY=0 in CI/production (hooks may fail when devDependencies not installed)
• Using deprecated shebang #!/usr/bin/env sh and husky.sh sourcing in hook files (will fail in v10)
• Using commitlint config with .js extension on Node v24 (use .mjs instead)

Gotchas & Edge Cases:

• git commit --no-verify bypasses hooks entirely - emergency escape hatch, not for regular use
• lint-staged function syntax () => "tsc --noEmit" runs on ALL files, not just staged
• Hook file names must match Git's exact names (pre-commit, commit-msg) - no extensions, case-sensitive
• HUSKY=2 enables debug mode (replaces deprecated HUSKY_DEBUG=1)
• commitlint v20+ is ESM-native - .mjs config extension avoids module loading issues
• Windows users need to escape $1 in commit-msg hook
• In monorepos, prepare script must navigate to repo root before running husky
• ~/.huskyrc support will be removed in v10 - migrate to .config/husky/init.sh
• lint-staged v16 uses picomatch (not micromatch) - glob patterns may differ slightly

</red_flags>

---

<critical_reminders>

CRITICAL REMINDERS

All code must follow project conventions in CLAUDE.md

(You MUST use "prepare": "husky" in package.json - NOT the deprecated "husky install")

(You MUST only lint staged files via lint-staged - NEVER lint the entire codebase in pre-commit)

(You MUST set HUSKY=0 in CI/production environments to disable hooks)

(You MUST use plain hook files in .husky/ directory - NO shebang lines or husky.sh sourcing in v9)

(You MUST keep pre-commit hooks under 10 seconds - move slow tasks to pre-push or CI)

Failure to follow these rules will cause slow commits, broken CI builds, and deprecated hook patterns that will fail in Husky v10.

</critical_reminders>