Hook Development

Adds event-driven automation to Claude Code through hooks that respond to tool use, session events, and user prompts. The prompt-based hooks are the standout feature here: instead of writing bash scripts to validate operations, you give Claude natural language criteria and let it reason about whether a file write is safe or a task is actually complete. You can still use command hooks for deterministic checks, but the LLM-driven approach handles edge cases better and requires less maintenance. Most useful for enforcing policies (blocking dangerous commands, validating completeness before stopping) and loading project context automatically. Hook configuration goes in your plugin's hooks.json, and you get proper input/output formats with permission decisions and system messages.