hook-development

To create event-driven hooks for Claude Code plugins, you define JSON decision outputs and scripts that approve, deny, modify, or augment tool usage. This Skill provides structured guidance for implementing prompt-based and command hooks across session lifecycle events.

A PreToolUse hook validates tool usage by intercepting actions before execution to approve or deny them. You can apply command checks for deterministic validation or LLM prompts for context-aware analysis, returning JSON decisions to enforce safety policies.

Yes, you can manage session lifecycle and state using SessionStart, SessionEnd, and SubagentStop hooks. These event-driven patterns allow you to load context, persist environment variables via CLAUDE_ENV_FILE, and automate workflows during session transitions.

Yes, this approach supports timeout and parallel execution handling for hooks. It includes schema validation, linting, and test harnesses alongside best practices to ensure your event-driven plugin hooks execute reliably without blocking workflows.

Yes, you need jq installed as a dependency to validate and run these Claude Code hooks. The Skill utilizes jq for processing JSON decision outputs and executing schema validation within the provided test utilities and scripts.

The best way to enforce safe file writes is by combining a PreToolUse command hook for quick path traversal checks with a prompt hook for deeper context analysis. This returns an approve or deny JSON decision before the tool executes.