Claude Code Hooks: The Missing Manual

Everyone talks about prompting. Nobody talks about hooks.

Hooks are shell commands that run in response to Claude Code events — before a tool executes, after a tool completes, or when a session starts. They're the programmable layer between you and the agent.

Why hooks matter

The default Claude Code experience is good. But hooks make it yours. They're how you:

• Prevent mistakes before they happen
• Enforce project conventions automatically
• Inject context at the right moment
• Build guardrails that don't slow you down

A simple example

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "command": "echo 'Use python3, not python' && [[ \"$TOOL_INPUT\" != *'python '* ]]"
      }
    ]
  }
}

This one line saves me from a mistake I make weekly on macOS where python doesn't exist.

The patterns that matter

After months of daily usage, three hook patterns emerged as genuinely useful:

1. Safety nets — Block destructive commands (rm -rf, git push --force)
2. Context injection — Feed project state into the session at startup
3. Convention enforcement — Ensure output matches your standards

Everything else is over-engineering. Start with these three.

What's next

I'll do a deep dive on each pattern with real examples from my workflow. The key insight: hooks are most powerful when they're invisible. The best hook is one you forget exists because it just works.