Walkthrough

Hooks: Pre and Post Tool

Run a script before or after each tool call. Lint on every edit; auto-test on commit.

Hooks let you wedge custom logic into the agent's loop. PreToolUse fires before the agent calls a tool — you can rewrite, allow, or block. PostToolUse fires after — perfect for running formatters, type-checks, or tests automatically. Other events cover the rest of the lifecycle (SessionStart, UserPromptSubmit, Stop), but the Pre/PostToolUse pair does most of the real work.

Configured in .claude/settings.json. The biggest win: post-edit linting. Every time the agent edits a file, your linter runs; if it fails, the agent sees the failure and self-corrects. You never see the broken state.

Steps · 0 / 3 done

Open the settings file
Create or open the project settings.
```
code .claude/settings.json
```
VerifyAn empty or existing settings.json opens.
Add a post-edit format hook
Run prettier on every Edit/Write to keep code consistent without asking the agent.
```
{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [{ "type": "command", "command": "npx prettier --write" }]
    }]
  }
}
```
VerifyAfter the agent edits a file, you see prettier run in the tool stream.
Add a pre-bash safety hook
Block the agent from running risky commands without asking. The hook intercepts and you decide.
```
{
  "hooks": {
    "PreToolUse": [{
      "matcher": "Bash",
      "hooks": [{ "type": "command", "command": "./scripts/audit-bash.sh" }]
    }]
  }
}
```
VerifyBash calls now route through your script. Add a deny list for things like `rm -rf /` and `git push --force`.

Check your understanding

Q1. Where do hooks live?

· Tick off the 3 step(s) above.

· Score 100% on the quiz.

Hooks: Pre and Post Tool

Open the settings file

Add a post-edit format hook

Add a pre-bash safety hook