Claude Code Hooks (Voice)

Why it's ranked here

The breadth is compelling: the handler covers every documented hook listed by the project, including agent-specific behavior and a special cue for Git commits. Cross-platform playback and defensive failure handling make it practical, though setup prerequisites and audio-player dependencies add friction.

What's good

Hook coverage extends beyond basic session and tool events to permissions, compaction, worktrees, tasks, configuration changes, and failures. Local settings override defaults without modifying shared configuration. Missing players or sounds fail quietly, while sound names receive a directory-traversal check.

Tradeoffs

Linux needs a supported external audio player. Windows playback is limited to WAV files, while the interactive lifecycle demo specifically requires macOS audio playback. Logging is enabled by default and records hook type, timestamp, and the last assistant message unless disabled.

How to use it well

It suits developers who run long or parallel Claude Code sessions and want ambient awareness without watching the terminal continuously. Disable noisy events locally and retain cues for permissions, failures, task completion, or session boundaries. It does not provide speech input, transcription, or a general notification service.

Observed

License

MIT License

Primary implementation language

Python

Install surface

Repository scripts and configuration with separate macOS, Linux, and Windows installation guides

Primary interface

Command-based Claude Code hook handlers receiving event data through standard input

Platform support

macOS, Linux, and Windows playback paths are implemented

Configuration

Local JSON configuration can override defaults and disable individual hooks or logging

Additional interface

A local HTTP demo exposes state retrieval and reset endpoints

MCP configuration

The repository configures the external @blizzy/mcp-elicit server through npx