What is the difference between exit code 0 and 2 in Claude Code hooks?

Exit code 0 indicates success, allowing Claude to process JSON or text output from stdout. Exit code 2 acts as a blocking error, where the content of stderr is sent back to Claude as a failure message.

What common fields are provided to every hook event?

Every hook receives a JSON object via stdin containing the session_id, transcript_path, current working directory (cwd), permission_mode, and the hook_event_name.

Can a hook modify a command before Claude executes it?

Yes, by returning an 'updatedInput' object within the 'hookSpecificOutput' of a PreToolUse or PermissionRequest hook, you can modify fields like the command string or file path.

How can I block a tool execution using a hook?

You can block tool execution by using the PreToolUse event and returning a JSON object with the hookSpecificOutput.permissionDecision field set to 'deny'.

Claude Code Hooks I/O API Reference

Name: Claude Code Hooks I/O API Reference
Author: Jamie-BitFlight

byJamie-BitFlight

•

Learning & Documentation

Provides comprehensive JSON schemas and I/O specifications for building and managing Claude Code hook scripts.

This skill serves as the definitive technical reference for developers building custom integrations with Claude Code's hook system. It details the exact JSON structures received via stdin for every major event—including PreToolUse, PostToolUse, and SessionStart—while providing the corresponding stdout schemas required to control Claude's behavior. Developers can use this reference to implement automated security gates, modify tool inputs dynamically, inject custom context, or manage subagent lifecycles with precision, ensuring seamless communication between external scripts and the Claude environment.

Key Features

01Comprehensive JSON schemas for all hook event types via stdin

0226 GitHub stars

03Standardized stdout response formats for controlling tool permissions

04Patterns for context injection during session start and prompt submission

05Detailed exit code mapping for blocking and non-blocking error handling

06Support for dynamic tool input modification using updatedInput

Use Cases

01Building audit logs and notification systems for subagent activities

02Implementing automated security policies that block sensitive shell commands

03Injecting custom project metadata or environment variables into the AI context

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jamie-bitflight/claude_skills hooks-io-api

For use in Claude.ai and ChatGPT

Download Skill