Hooks

This is a reference guide for Claude's hook system, which lets you intercept and modify agent behavior at specific points like before tool execution, after prompts, or during session starts. You write hooks as shell scripts that call TypeScript handlers via tsx, reading JSON input from stdin and outputting whether to continue or block. It's most useful when you need guardrails on what tools Claude can use, want to inject context at session boundaries, or need to log agent actions. The pattern is clean but requires understanding the event lifecycle. If you're building a Claude workflow that needs programmatic control over agent decisions rather than just prompt engineering, this is how you do it.

skills.sh installs: 509
GitHub stars: 3.9K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add parcadei/continuous-claude-v3 --skill hooks --agent claude-code

Installs into .claude/skills of the current project.

Files

Hook Development Rules

When working with files in .claude/hooks/:

Pattern

Shell wrapper (.sh) → TypeScript (.ts) via npx tsx

Shell Wrapper Template

#!/bin/bash
set -e
cd "$CLAUDE_PROJECT_DIR/.claude/hooks"
cat | npx tsx <handler>.ts

TypeScript Handler Pattern

interface HookInput {
  // Event-specific fields
}

async function main() {
  const input: HookInput = JSON.parse(await readStdin());

  // Process input

  const output = {
    result: 'continue',  // or 'block'
    message: 'Optional system reminder'
  };

  console.log(JSON.stringify(output));
}

Hook Events

PreToolUse - Before tool execution (can block)
PostToolUse - After tool execution
UserPromptSubmit - Before processing user prompt
PreCompact - Before context compaction
SessionStart - On session start/resume/compact
Stop - When agent finishes

Testing

Test hooks manually:

echo '{"type": "resume"}' | .claude/hooks/session-start-continuity.sh

SKILL.md on GitHub

Details

Repository: github.com/parcadei/continuous-claude-v3
Category: Frontend Development
First seen: Jun 3, 2026