creating-claude-hooks

Claude Code hooks are executable scripts that trigger on session-start, user-prompt-submit, tool-call, and assistant-response events. Write hooks as executables with a proper shebang, accept JSON input via stdin, and output text to stdout with appropriate exit codes to signal success or failure.

Hooks must be executable files with a shebang line (e.g., #!/bin/bash), receive JSON event data on stdin, write output to stdout, and return exit code 0 for success or non-zero for errors. This stdin/stdout pattern enables Claude Code to pass event context and capture hook results consistently.

Package hooks using PRPM format with proper directory structure and metadata. Include your hook executable, schema definitions for validation, and security configurations. PRPM packaging standardizes hook distribution, enables dependency resolution, and ensures hooks meet security and format requirements before publication.

Hooks must pass schema validation to ensure correct input/output structure and comply with security requirements specified in PRPM packaging. These requirements prevent malformed hooks from executing and protect the Claude Code environment from unvalidated or unsafe hook behavior.

Yes. Hook events like tool-call and assistant-response let you trigger automation in response to Claude Code actions. For example, create a tool-call hook to run code formatters on modified files or an assistant-response hook to post-process outputs automatically.

No prior automation experience is strictly required, but familiarity with shell scripting or executable development helps. The Skill covers hook format, event handling, and I/O conventions step-by-step, making it accessible for developers new to hook creation.