What is the primary advantage of using Hooks over Skills in Claude Code?

Hooks offer a 100% execution guarantee because they are event-based and do not rely on Claude's judgment, whereas Skills or Agents are triggered based on Claude's decision-making, which has variable reliability.

Can Claude Code hooks modify tool inputs before they execute?

Yes, specifically during the PreToolUse and PermissionRequest events, a hook can return an 'updatedInput' to modify the command or parameters before they are processed.

How is session data passed to a hook script?

Session data, including session_id, cwd, and transcript_path, is passed to the hook script via stdin in JSON format. Hooks must read stdin (e.g., using cat or jq) to access this data.

Which hook event is best for injecting project-specific context?

The UserPromptSubmit event is ideal for context injection because any output to stdout during this event is automatically injected into Claude's prompt context.

What does Exit Code 2 signify in the hook system?

Exit Code 2 acts as a 'Block' or 'Special Action' signal. Its behavior depends on the event; for example, in PreToolUse it blocks execution, while in the Stop event it forces Claude to continue working.

Claude Code Hook Architect Reference

Name: Claude Code Hook Architect Reference
Author: chkim-su

bychkim-su

0•

開発者ツール

Provides a comprehensive technical reference for Claude Code's hook system to build deterministic, event-driven automation workflows.

This skill acts as the definitive documentation and architectural guide for the Claude Code Hook system, a powerful mechanism that guarantees 100% execution reliability by bypassing AI judgment. It detail five essential hook roles—Gate, Side Effect, State Manager, External Integrator, and Context Injector—enabling developers to build robust pipelines for security scanning, automated linting, and state management. With deep dives into 10 event types, 22 universal implementation patterns, and precise exit code behaviors, this reference empowers users to extend Claude's capabilities with predictable, programmatic control over every step of the AI interaction lifecycle.

主な機能

010 GitHub stars

02Detailed implementation patterns for Gate, Side Effect, and Context Injector roles

03Technical specifications for stdin JSON data structures and session metadata passing

04Comprehensive guide to 10+ hook events including SessionStart, PreToolUse, and Stop

05Best practices for infinite loop prevention, path normalization, and dry-run enforcement

06Advanced workflow control using Exit Code 2 for blocking or forcing continuations

ユースケース

01Implementing auto-formatters and linters that execute immediately after tool use

02Creating automated security gates to block dangerous commands or secret leaks

03Developing custom state managers to track complex multi-step developer workflows

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add chkim-su/forge-editor hook-capabilities

For use in Claude.ai and ChatGPT

Download Skill