What is the difference between a command hook and a prompt hook?

Command hooks execute standard shell scripts for deterministic tasks, while prompt hooks use LLM evaluation to make context-aware decisions about continuing or blocking a task.

Can I use environment variables in my hooks?

Yes, Claude provides standard variables like CLAUDE_PROJECT_DIR, and you can persist custom variables using the CLAUDE_ENV_FILE during SessionStart events.

Where are hook configurations stored?

Hooks are configured in settings.json files located in either the global ~/.claude/ directory or the project-specific .claude/ directory.

How do I block a Claude action using a hook?

To block an action, your hook script should return an exit code of 2. Any message sent to stderr will be displayed to Claude as the reason for the block.

Claude Hook Development

Name: Claude Hook Development
Author: cmtkdot

bycmtkdot

0•

开发者工具

Automates and controls Claude's workflow through deterministic scripts and LLM-evaluated lifecycle hooks.

The Hook Development skill provides a comprehensive framework for building, configuring, and debugging Claude Code hooks, enabling developers to automate tasks at specific lifecycle events. Unlike standard instructions, hooks are deterministic, allowing for strict control over events such as tool execution, session starts, and prompt submissions. This skill guides users through creating bash-based command hooks and context-aware prompt hooks, implementing security guardrails to block dangerous commands, and optimizing workflows with automated linting or auto-approval patterns.

主要功能

01Scaffolding for .claude/settings.json and project-specific hook structures

02Exit code management for blocking or allowing AI actions with custom feedback

03Deterministic control over lifecycle events like PreToolUse and SessionStart

04Support for both Bash command hooks and LLM-evaluated prompt hooks

050 GitHub stars

06Regex-based tool matchers for granular execution control

使用场景

01Implementing security guardrails to block dangerous commands like forced git pushes

02Automatically running linters or test suites after Claude edits a source file

03Auto-approving modifications to safe file types like documentation or markdown

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add cmtkdot/claude-development hook-development

For use in Claude.ai and ChatGPT

Download Skill