Claude Code Hooks: 6 Copy-Paste Examples (Block rm -rf, Protect .env, Auto-Format)

Published Oct 15, 2026 · 7 min read · Checked against the official hooks reference

Hooks are the only part of Claude Code that are guaranteed to run. CLAUDE.md is advice and skills are instructions. A hook runs every time its event fires, whatever the model decides. If a rule must never be broken, it belongs in a hook.

How hooks work in 30 seconds

• You define them in .claude/settings.json (project, committed), .claude/settings.local.json (just you) or ~/.claude/settings.json (all projects).
• Structure: event (e.g. PreToolUse) → matcher (tool name such as Bash or Edit|Write) → handlers (usually "type": "command").
• Your script receives JSON on stdin with tool_name and tool_input.
• Exit 2 blocks the action (on events that can block), and stderr is shown to Claude. Exit 0 means no objection. Other exit codes are non-blocking errors. Exit 1 does not block, which catches a lot of people out.
• Use ${CLAUDE_PROJECT_DIR} in paths, and run /hooks to see what's loaded.

1. Block destructive shell commands

#!/usr/bin/env bash
# .claude/hooks/block-dangerous.sh
cmd=$(jq -r '.tool_input.command // ""')
if echo "$cmd" | grep -Eq 'rm -rf (/|~|\*)|git push (-f|--force)|git reset --hard|mkfs|dd if='; then
  echo "Blocked: '$cmd' looks destructive. Ask the user to run it manually." >&2
  exit 2
fi
exit 0

{ "hooks": { "PreToolUse": [ { "matcher": "Bash", "hooks": [
  { "type": "command", "command": "\"${CLAUDE_PROJECT_DIR}\"/.claude/hooks/block-dangerous.sh" } ] } ] } }

2. Protect .env, lockfiles and .git

#!/usr/bin/env bash
# .claude/hooks/protect-files.sh  (matcher: "Edit|Write")
path=$(jq -r '.tool_input.file_path // ""')
case "$path" in
  *.env|*.env.*|*/.git/*|*package-lock.json|*pnpm-lock.yaml|*yarn.lock)
    echo "Blocked: $path is protected." >&2; exit 2;;
esac
exit 0

3. Auto-format after every edit

{ "hooks": { "PostToolUse": [ { "matcher": "Edit|Write", "hooks": [
  { "type": "command", "command": "jq -r '.tool_input.file_path' | xargs npx prettier --write", "timeout": 30 } ] } ] } }

PostToolUse can't block, because the edit already happened. Its job is cleanup.

4. Don't let Claude stop until tests pass

#!/usr/bin/env bash
# Stop hook: exit 2 sends Claude back to work with this message
if ! npm test --silent >/tmp/cc-test.log 2>&1; then
  echo "Tests are failing. Fix them before finishing:" >&2; tail -20 /tmp/cc-test.log >&2; exit 2
fi
exit 0

Register it under "Stop" (no matcher). Guard against infinite loops: the input includes stop_hook_active, so exit 0 when it's true.

5. Desktop ping when Claude needs you

{ "hooks": { "Notification": [ { "hooks": [
  { "type": "command", "command": "notify-send 'Claude Code' 'Needs your input'" } ] } ] } }

On macOS, use osascript -e 'display notification "Needs input" with title "Claude Code"'.

6. Inject live context at session start

{ "hooks": { "SessionStart": [ { "matcher": "startup", "hooks": [
  { "type": "command", "command": "echo \"Branch: $(git branch --show-current). Open TODOs: $(grep -rc TODO src | awk -F: '{s+=$2} END {print s}')\"" } ] } ] } }

Plain stdout from a SessionStart hook is added to Claude's context.

Gotchas

• Make scripts executable (chmod +x) and install jq.
• A regex blocklist is a seatbelt, not a sandbox. Pair it with permission deny rules.
• Print only JSON to stdout if you return JSON. Shell-profile noise breaks parsing.
• Debug with claude --debug.

Reference: code.claude.com/docs/en/hooks