Hook Development

Event-driven hooks intercept tool usage through PreToolUse events, allowing you to inspect and block dangerous operations like `rm -rf` or privilege escalation before execution. Add a prompt-based hook to your plugin's hooks.json that evaluates each command against your safety policy.

Hook Development supports automation across six event types: PreToolUse and PostToolUse for tool validation, SessionStart and UserPromptSubmit for context initialization, and Stop and SubagentStop for cleanup. Each event triggers custom scripts or prompt-based logic to enforce policies or load environment variables.

Yes. Use SessionStart hooks to execute detection logic and populate environment variables via CLAUDE_ENV_FILE. This enables context-aware automation that adapts your Claude Code environment based on the project structure detected at session initialization.

Hook Development includes validation, testing, and linting tooling that ensures consistent, safe hook implementations. These tools verify your hooks.json configuration, test event triggers, and enforce standardized outputs before deployment to your Claude Code plugin.

Yes. jq is a required dependency for parsing and manipulating JSON within hook configurations and command-based validations. It enables complex filtering and transformation logic in your event-driven automation workflows.

Prompt-based hooks use Claude's reasoning to evaluate tool usage against natural-language policies, while command-based hooks execute shell scripts for direct, deterministic validation. Both are supported for PreToolUse events to suit different safety requirements and performance needs.