prompting-toolkit

Prompting Toolkit

A comprehensive prompting ecosystem with real-time enhancement and advanced strategy library

📺 Assets & Media

Explainer videos and architecture diagrams are available in the assets/ directory:

• Explainer Video: Click image to watch PBS-structured explainer video (2:15) - Problem → Behavior → Solution
• Architecture Diagram: See assets/diagrams/architecture.md for system design overview
• Integration Examples: See assets/examples/ for hook and framework integration patterns

Note: Media assets are generated using NotebookLM and Claude Code's built-in diagramming tools.

🎯 Two Installation Modes

This package supports two installation modes depending on your Claude Code setup:

Mode A: Router-Based Installation (Recommended for Advanced Users)

If you use a custom hook router system (like UserPromptSubmit_router.py), install via the standard plugin pattern:

# 1. Clone or copy this package to P:/packages/
git clone <repo-url> P:/packages/prompting-toolkit

# 2. The .claude-plugin directory will be auto-discovered by Claude Code
# 3. Your router (P:/.claude/hooks/UserPromptSubmit_router.py) will handle the hooks

How it works:

• .claude-plugin/plugin.json declares this as a Claude Code plugin
• hooks.json routes UserPromptSubmit events to your custom router
• Your router consolidates multiple hooks and manages execution order
• This is the hybrid approach: official plugin structure + custom router integration

Why use this mode:

• ✅ Centralized hook management in your router
• ✅ Fine-grained control over hook priorities and execution order
• ✅ Easy to debug and test hooks via router
• ✅ Can add custom preprocessing/postprocessing in router

Mode B: Default Claude Code Installation (Standard Claude Code Users)

If you're using default Claude Code functionality without custom routers:

# 1. Copy the hook package directly to Claude's hooks directory
cp -r P:/packages/prompting-toolkit/packages/hook ~/.claude/hooks/_packages/prompt-hook

# 2. Enable in Claude Code settings.json
claude config set hooks.UserPromptSubmit.prompt-hook.enabled true

Or add to your ~/.claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": ".*",
        "hooks": [
          {
            "type": "command",
            "command": "python ~/.claude/hooks/_packages/prompt-hook/hook/prompt_enhancement.py"
          }
        ]
      }
    ]
  }
}

How it works:

• Direct hook execution without router layer
• Simpler setup for standard Claude Code installations
• Each hook is registered independently in settings.json

Why use this mode:

• ✅ No custom router required
• ✅ Standard Claude Code plugin pattern
• ✅ Easier for users without advanced hook setups
• ✅ Works out-of-the-box with default Claude Code

🔧 Configuration

Environment Variables

Both installation modes support these environment variables:

# Enable/disable prompt enhancement
export PROMPT_ENHANCEMENT_ENABLED=true

# Enable interactive choice UI (ask user before applying enhancement)
export PROMPT_CHOICE_ENABLED=true

# Debug mode (verbose logging)
export PROMPT_ENHANCEMENT_DEBUG=false

Add these to your P:/.claude/settings.json under the "env" section:

{
  "env": {
    "PROMPT_ENHANCEMENT_ENABLED": "true",
    "PROMPT_CHOICE_ENABLED": "true"
  }
}

📊 Architecture

This monorepo contains two complementary packages:

packages/hook (formerly prompt-enhancement)

User-facing prompt improver that runs automatically in Claude Code.

• ✨ Noise Cleaning - Removes terminal artifacts automatically
• 🎯 Complexity Analysis - Categorizes prompts (simple/moderate/complex/expert)
• 🏷️ Domain Detection - Identifies context (security, testing, database, frontend)
• 🔀 Choice UI - Interactive approval: enhanced vs. original prompt
• 🔒 Multi-terminal Safe - Isolated state per terminal session

Perfect for: Claude Code users who want automatic prompt improvement without leaving their workflow.

packages/framework (formerly prompting-framework)

Developer library with 23+ prompting strategies and optimization algorithms.