Can hooks prevent Claude from executing a specific tool?

Yes, PreToolUse hooks are blocking-capable. By returning a JSON object with 'decision': 'block', you can stop the agent from proceeding with that specific tool execution.

What is the difference between command and prompt hooks?

Command hooks execute shell scripts for deterministic tasks like logging or formatting, while prompt hooks use the LLM to make complex, reasoning-based decisions like evaluating if a command is safe.

Where are the hook configuration files located?

For Claude Code, hooks are defined in .claude/hooks.json at the project level or ~/.claude/hooks.json for global user settings.

How do I prevent hooks from creating infinite loops?

When configuring Stop hooks, always check the 'stop_hook_active' flag in the input JSON to ensure the hook logic does not trigger itself recursively.

Agent Event Hooks Manager

Name: Agent Event Hooks Manager
Author: utahdws

byutahdws

生産性とワークフロー

Automates agent workflows and enforces safety protocols through event-driven shell commands and LLM-evaluated prompts.

概要

Agent Event Hooks Manager provides a robust framework for extending Claude Code and Gemini CLI with custom automation. It allows developers to intercept agent actions at specific lifecycle events—such as before tool execution or at session start—to perform validations, inject context, or trigger external workflows like code formatting and desktop notifications. By leveraging both shell-based commands and context-aware LLM prompts, this skill empowers teams to implement project-specific guardrails and personalized developer experiences without modifying the core agent logic.

主な機能

Dual-mode execution via shell commands or reasoning-based LLM prompts
0 GitHub stars
Blocking mechanisms to prevent destructive actions or enforce compliance
Regex-based matchers for granular control over tool interception
Lifecycle triggers for tool usage, prompt submission, and session state
Multi-agent support for Claude Code and Gemini CLI event mapping

ユースケース

Blocking destructive CLI operations like force pushes or unauthorized deletions
Injecting specific project context or sprint goals automatically at session start
Triggering automated linting and formatting after agent-led code modifications

概要

主な機能

Dual-mode execution via shell commands or reasoning-based LLM prompts
0 GitHub stars
Blocking mechanisms to prevent destructive actions or enforce compliance
Regex-based matchers for granular control over tool interception
Lifecycle triggers for tool usage, prompt submission, and session state
Multi-agent support for Claude Code and Gemini CLI event mapping

ユースケース

Blocking destructive CLI operations like force pushes or unauthorized deletions
Injecting specific project context or sprint goals automatically at session start
Triggering automated linting and formatting after agent-led code modifications