Hooks, part one: events and configuration

Hooks run your own logic on lifecycle events — deterministically, every time the event fires. The main events:

Event	When	Use for
PreToolUse	Before a tool runs	Validation, approve/deny/modify
PostToolUse	After a tool completes	Feedback, formatting, logging
UserPromptSubmit	You submit a prompt	Add context, validate
Stop / SubagentStop	Agent considers stopping	Completeness checks
SessionStart / SessionEnd	Session boundaries	Load context / cleanup
PreCompact	Before compaction	Preserve critical info
Notification	Claude notifies you	Logging, reactions

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

Matchers select which tools trigger the hook: exact ("Write"), alternation ("Read|Write|Edit"), wildcard ("*"), or regex against MCP tools ("mcp__.*__delete.*"). Hooks receive JSON on stdin (tool_name, tool_input, cwd…) and speak through exit codes: 0 = success, 2 = blocking error whose stderr is fed back to Claude.

Check your understanding

Question 1Which exit code makes a hook block the action and feed stderr back to Claude?

Question 2You edited hooks.json mid-session. What must you do?

Question 3Which matcher fires on every MCP tool?