John Lindquist's TypeScript Hooks for Claude Code

Table of content

John Lindquist co-founded egghead.io in 2013 and invented the bite-sized screencast format that taught millions of developers AngularJS, React, and TypeScript. After 16 years of teaching developers, he now works on AI developer experience at Vercel and has spent over 1 billion tokens using Claude Code.

His claude-hooks project solves a specific problem: Claude Code hooks are powerful but writing them in raw shell scripts means no autocomplete, no type checking, and constant trips to the documentation to remember payload structures.

Background

Co-founded egghead.io in 2013, turning free AngularJS screencasts into a developer education platform
Previously Technical Architect at Roundarch, building apps for the Air Force, HBO, and Bloomberg
WebStorm evangelist at JetBrains before going full-time on egghead
Created Script Kit, an automation tool with 4.2k GitHub stars
Currently AI DX at Vercel

GitHub | Twitter | Site

Type-Safe Hooks

The claude-hooks project gives you TypeScript definitions for all hook events. Instead of guessing what fields exist on a PreToolUse payload, you get IntelliSense:

npx claude-hooks

This scaffolds a .claude/hooks/ directory with typed payloads:

// .claude/hooks/index.ts
import { PreToolUse, PostToolUse, Notification, Stop } from './lib'

export async function preToolUse(payload: PreToolUse) {
  // payload.tool_name, payload.tool_input fully typed
  if (payload.tool_name === 'write_file') {
    // Block writes to production files
    if (payload.tool_input.path.includes('production')) {
      return { action: 'block', reason: 'Cannot modify production files' }
    }
  }
}

export async function postToolUse(payload: PostToolUse) {
  // Run formatters after file edits
  if (payload.tool_name === 'write_file') {
    await formatFile(payload.tool_input.path)
  }
}

The type definitions cover all eight lifecycle events: UserPromptSubmit, PreToolUse, PostToolUse, Notification, Stop, and the session variants.

Hook Events

Event	When It Fires	Use Case
UserPromptSubmit	Before Claude sees your prompt	Validate or enhance prompts
PreToolUse	Before any tool execution	Block dangerous operations
PostToolUse	After tool completion	Format code, run tests
Notification	When Claude sends notifications	Custom alerts (Slack, native)
Stop	When Claude finishes responding	Session cleanup, logging

Beyond Hooks

John also created mdflow, a CLI that runs Markdown files as executable prompts against Claude, Codex, Gemini, or Copilot. Write your prompt once in .md, run it against any backend:

mdflow run prompt.md --backend claude
mdflow run prompt.md --backend codex

His Claude Code Power-User workshop has trained 333 developers across seven cohorts, covering hooks, the SDK, sub-agents, and MCP integrations.

Key Takeaways

Principle	Implementation
Type everything	Use claude-hooks for typed payloads instead of raw JSON
Automate formatting	PostToolUse hooks run prettier/eslint after every edit
Block mistakes early	PreToolUse hooks prevent writes to sensitive paths
Learn from volume	1 billion tokens reveals patterns you miss in casual use