this is part 3 of

Claude Code Tips VIDEO 3: ADVANCED BEGINNER Claude.md Model Choice Plan Mode Escape Stop Screenshots INTERMEDIATE bash mode ! /context /compact /clear /resume ADVANCED Hooks Subagents Custom Cmd Permissions MCP Claude Code - Part 3 Hooks this is part 3 of How to Master Claude Code from intro to advanced Today we're going to talk about an advanced technique called hooks we're gonna go through what it is, what it's good for what a hook looks like and then do an example to show you how to put in a hook Hooks Custom actions that automatically run at specific stages in Claude Code. Trigger before/after tool use, on permission requests, when calls complete, or at session start/end. Automate initialization, logging, checks, and cleanup tasks. so hooks are customizable things that Claude Code can do for you at different points of the life cycle so some common points within a life cycle include Event Trigger Use Case PreToolUse Before tool Block risky commands, validate input PostToolUse After tool Auto-format code, run tests, log changes PermissionRequest Permission ask Alert sound, custom approval flow Stop Response done Play notification, send Slack message SessionStart Session begins Load context, initialize environment SessionEnd Session closes Cleanup, final logging so some common points within a life cycle include pre tool used - before you use the tool post tool use - after you use the tool permission request which is when permission is requested of you Stop which is when Claude has finished a task session start when you start Claude code session start when you start Claude code and session end when you're ending a session with Claude code these are all parts of the lifecycle where you can ask Claude to do something specific when this event is triggered for example when a permission is requested for example when a permission is requested you can ask Claude to notify you or send you a sound alert or for example at a session start you can ask Claude Code to load context or initialize the environment another example is when a session ends you can automatically get Claude Code to do some cleanup and some logging so think of it as when we get to this point or this event Claude Code will automatically do these things for you now let's go through what a hook looks like and this is what we're going to put in at the end as an example this hook is going to do two things it will play a chirping sound when it's asking me for permission on something and then it will play a meow sound when it is finished doing a command I don't know about you guys but I find myself constantly checking the screen in order to find out if Claude Code is waiting for me for permission or waiting for me for the next move so I wanted it to play these sounds at these stages now let's look through each line "hooks" -- Root configuration key This is the parent object that contains all your hook definitions. Everything else nests inside this. OTHER VARIATIONS: "hooks" (always lowercase, always plural) Located in: ~/.claude/settings.json at the very top this says hook this is the root configuration key this kind of just says this whole entire object is a hook the next one is permission request "PermissionRequest" -- Event type This specifies WHEN the hook fires. When Claude asks for permission, this hook triggers. OTHER VARIATIONS: "Stop" -- When Claude finishes "PreToolUse" -- Before any tool runs "PostToolUse" -- After a tool completes "SessionStart" -- When you start "SessionEnd" -- When you exit "Notification" -- When Claude needs attention this is an event type so this is I mentioned about the different parts of the lifecycle you can change it to stop pre tool use, post tool use session start, session end, notification any point of development the next one is a matcher "matcher" -- Filter condition Specifies WHICH tools or conditions trigger the hook. Empty string "" means match everything. You can narrow it down to specific tools. OTHER VARIATIONS: "Bash" -- only Bash tool "Edit|Write" -- either Edit OR Write "--" -- match all (wildcard) "^re" -- regex patterns work too this is a filtering condition so you can specify the tool or condition that it must match for me there's no specific condition anytime it's asking for permission request it should be playing this sound "hooks" -- Handler array Array of actions that execute when the event fires and matcher matches. Each object is one command/request/prompt. Can have multiple handlers Handlers run in sequence Each handler needs "type" and "command"/"uri"/"prompt"/"agent" the next one is the start of a hook array so you can actually get it to do multiple actions under this condition notice here I only have one thing in my array because I'm only asking it to play one sound if I say you want it to play a sound and have it pop up on your screen then you can put two of those items in the same array "type" -- Handler type Specifies HOW the handler executes. "command" means run a shell command. The type determines what the next key should be. OTHER VARIATIONS: "command" + next key: "command" "http" + next key: "uri" "prompt" + next key: "prompt" "agent" + next key: "agent" for this example this is of type command which means it will run a shell command "command" -- The actual shell command The shell command that runs when all conditions match. Also access to environment variables like $CLAUDE_TOOL_INPUT_FILE_PATH, $CLAUDE_SELECT_DIR and $CLAUDE_TOOL_INPUT_FILE_PATH. OTHER VARIATIONS: "afplay ~/claude/sounds/meow.mp3 & (play sound)" npx prettier --write $FILE (format code) npm test -- --findRelatedTests (run tests) echo "msg" >> log.txt (log event) and then the actual command it will run is it will go into this folder and play this MP3 sound called chirp so this is when permission is requested now the same thing happens here this is when instead of permission requested this is the event stop so when Claude Code has finished running or stopped running play the sound meow so that's the basic structure there are a few other optional parameters that you can put in there you can always ask Claude Code just to write it out for you let me show you how to put this chunk of code in so that it works so you're gonna open up your Claude folder and open a new folder here called sounds soundtracks in there and then open up your settings here you'll see there's already some text in here take your hook and paste it in here so you see this is my chirp and this is my meow save and close and that's it that's all you need now let's see an example of it working permission requested *chirp* I'm providing my approval Task Complete *meow* (the sound actually played a bit before the complete message got displayed) so that's basically how a hook works it's really good for anything you want to automatically run when you are at a point in the life cycle or at an event I hope that's helpful and I'll see you at the next one for another advanced technique in Claude Code