⚡ Prompt Refiner

A developer-native terminal UI that automatically refines your prompts using proven prompt engineering techniques — powered by your existing Claude Code subscription. No separate API key required.

Screenshot

Demo

┌─────────────────────────────────────────────────────────────────────┐
│  ⚡ Prompt Refiner   Model: [claude-sonnet-4-6 ▾]                   │
├────────────────────────┬────────────────────────────────────────────┤
│  YOUR PROMPT           │  REFINED PROMPT                            │
│                        │                                            │
│  write a function to   │  You are an expert software engineer.      │
│  parse json            │                                            │
│                        │  <instructions>                            │
│                        │  Write a function to parse JSON input.     │
│                        │  Think step by step before answering.      │
│                        │  </instructions>                           │
│                        │                                            │
│                        │  <output_format>                           │
│                        │  Provide a complete, documented function   │
│                        │  with type hints and error handling.       │
│                        │  </output_format>                          │
├────────────────────────┴────────────────────────────────────────────┤
│  ANNOTATIONS                                                        │
│  ► Role prompting — code generation task detected                   │
│  ► Chain-of-thought — improves accuracy on multi-step logic         │
│  ► XML structuring — delineates sections clearly for Claude         │
│  ► Output format — specifies expected structure and style           │
├─────────────────────────────────────────────────────────────────────┤
│  ^R Refine  y Accept  n Reject  e Edit+Refine  Tab Model  q Quit    │
└─────────────────────────────────────────────────────────────────────┘

Features

• Auto-detects task type — code, debug, explanation, or general
• Applies relevant techniques — role prompting, chain-of-thought, XML structuring, output format, constraint injection
• Annotates every change — explains why each technique was applied
• 3-action workflow — Accept (copies to clipboard), Reject, or Refine with comments
• Model switching — cycle between Haiku, Sonnet, and Opus with Tab
• Zero setup — uses your existing Claude Code login

Prerequisites

• Claude Code installed and logged in (claude login)
• Python 3.11+

Installation

Via pipx (recommended — isolated, no venv management):

pipx install claude-prompt-refiner

Via Homebrew:

brew tap richeshgupta/claude-prompt-refiner https://github.com/richeshgupta/claude-prompt-refiner
brew install claude-prompt-refiner

Via pip:

pip install claude-prompt-refiner

Usage

cpr                    # short alias
claude-prompt-refiner  # full command

Keybindings

Prompt Techniques Applied

Project Structure

├── main.py          # Entry point
├── app.py           # Textual TUI (layout, keybindings, workers)
├── engine.py        # RefinerEngine — prompt building + JSON parsing
├── claude.py        # Subprocess wrapper for claude CLI
├── techniques.py    # Task detection + technique catalogue
├── requirements.txt
└── tests/           # 17 unit tests

Running Tests

python -m pytest tests/ -v

Models

Contributing

See CONTRIBUTING.md.

License

MIT