Hooks — Make the Agent Obey Your Rules

You can ask the agent to run the formatter after every edit. Or you can guarantee it. Hooks are shell commands Claude Code runs automatically at defined points in its loop — turning "please remember" into "always."

The events you'll actually use

• PostToolUse — after a tool succeeds. Format or lint the file just edited.
• PreToolUse — before a tool runs. Inspect and block dangerous calls.
• UserPromptSubmit — inject context before the model sees your prompt.
• SessionStart / Stop — load context at the start, notify at the end.

Wiring one up

Hooks live under a hooks key in settings.json. Each event maps to matchers (a regex over the tool name) and commands:

{
  "hooks": {
    "PostToolUse": [
      { "matcher": "Edit|Write",
        "hooks": [{ "type": "command", "command": "npx prettier --write \"$CLAUDE_FILE_PATHS\"" }] }
    ]
  }
}

Exit codes are the control channel

• 0 — success; the agent continues.
• 2 — blocking error; the action is prevented and stderr is fed back so the agent can adjust.
• other non-zero — non-blocking error, logged.

Recipes that earn their keep

• Auto-format on edit — PostToolUse + Edit|Write.
• Block rm -rf / force-push — PreToolUse + Bash, exit 2 on match.
• Secret scan before write — PreToolUse + Write, block on detected keys.
• Desktop ping when input is needed — Notification.

Hooks enforce behavior. Next we package intent — reusable team workflows — as custom slash commands.