Hooks & Automation

What Are Hooks?

You’ve set up memory (CLAUDE.md), and you’ve created skills (SKILL.md). Now it’s time to add automation — things that happen without you asking.

Hooks are shell commands that run automatically at specific moments during a Claude Code session. They’re like setting up dominoes: when one thing happens, another thing happens automatically.

Here’s the analogy: imagine you have a doorbell camera. When someone approaches your door (the trigger), it automatically starts recording (the action). You don’t have to press a button — it just happens. Hooks work the same way.

The Four Hook Types

There are four moments when hooks can fire:

1. PreToolUse — Runs before Claude uses a tool (like editing a file or running a command). Perfect for blocking dangerous operations.

2. PostToolUse — Runs after Claude uses a tool. Great for automatic cleanup, like running a code formatter after every file edit.

3. SessionStart — Runs when you start a new Claude Code session. Useful for setting up environment variables or checking system status.

4. Notification — Runs when Claude sends a notification (like when a long task finishes). You can use this to trigger desktop alerts or sounds.

Think of it like a timeline:

Where Hooks Live

Hooks are configured in your settings.json file. You can edit this file directly, or use the /hooks command inside Claude Code to manage them through a menu.

The settings file lives at:

~/.claude/settings.json

Here’s what a hook looks like in that file:

This hook says: “Every time Claude finishes editing a file, automatically run the Prettier code formatter on that file.” You never have to remember to format your code — it just happens.

Pro Tip

The matcher field filters which tool triggers the hook. "Edit" means only file edits. "Bash" means only terminal commands. Leave it empty to match everything.

Blocking Dangerous Commands

One of the most powerful uses of PreToolUse hooks is blocking dangerous operations before they happen. Think of it as a safety net.

For example, you might want to prevent Claude from ever running rm -rf / (which deletes everything) or git push --force (which can destroy your team’s work):

When a PreToolUse hook exits with code 1 (an error), Claude’s action is blocked. It won’t run the dangerous command. Exit code 0 means “go ahead.”

Auto-Lint After Every Edit

Here’s a practical PostToolUse hook that runs ESLint (a code quality checker) after Claude edits any JavaScript file:

Every time Claude edits a .js file, ESLint automatically checks and fixes it. No extra steps, no forgetting.

Pro Tip

You can chain multiple PostToolUse hooks. For example: format with Prettier, then lint with ESLint, then run the relevant test file — all automatically after every edit.

SessionStart: Your Morning Briefing

SessionStart hooks run once when you open Claude Code. They’re perfect for setting up your environment or giving yourself a status report.

Here’s a hook that loads environment variables and checks your Git status:

You can get creative with this. Some people set up a “morning briefing” that shows:

• How many unread Slack messages they have
• What’s on their calendar today
• Which pull requests need review
• The status of their last deployment

All of that shows up automatically when they type claude.

Set up a PostToolUse hook in my settings.json that runs Prettier on any file after Claude edits it. Also add a PreToolUse hook that blocks any git push --force command.

Recurring Tasks with /loop

Sometimes you want Claude to repeat a task on a schedule — like checking a deployment every 5 minutes or monitoring a log file for errors.

The /loop command does exactly this:

This tells Claude to run that prompt every 5 minutes. It keeps going until you stop it. You can use any time interval: 30s (30 seconds), 5m (5 minutes), 1h (1 hour).

/loop 5m check the deployment at https://my-app.vercel.app and tell me if it returns any errors

/loop 2m read the last 20 lines of server.log and alert me if there are any ERROR or WARN entries

Scheduled Tasks

Beyond /loop (which runs during your session), you can set up tasks that run even when you’re not at your computer:

• Desktop scheduled tasks — Use your operating system’s built-in scheduler (cron on Mac/Linux, Task Scheduler on Windows) to run claude -p "your prompt" at specific times
• GitHub Actions — Run Claude Code in the cloud on a schedule, like checking for security vulnerabilities every night
• Cloud functions — Trigger Claude from AWS Lambda, Google Cloud Functions, or similar services

Create a GitHub Action that runs every night at midnight and uses Claude Code to scan the codebase for common security vulnerabilities. Have it open an issue if it finds anything.

Try It Yourself