Mastering Prompt Engineering: Building a CLI Command Translator with the ChatGPT API

The article introduces the official ChatGPT API and emphasizes the critical role of Prompt Engineering in transforming general-purpose AI into specialized productivity tools. It begins by outlining the prerequisites, including OpenAI account registration and API key generation, while highlighting the capabilities and cost-efficiency of the gpt-3.5-turbo model.

The core of the tutorial focuses on iterative prompt design using the OpenAI Playground. It details seven progressive steps: establishing the system role, enforcing direct output constraints, handling invalid inputs with fallback responses, formatting outputs as single-line commands, adding safety warnings for dangerous operations, utilizing few-shot examples to correct model behavior, and translating prompts to English for optimal performance.

Following the prompt design phase, the author outlines a modular implementation architecture using Node.js, TypeScript, and the OpenAI SDK. The system is divided into four components: an API wrapper for streamlined requests, a prompt generator that loads templates and injects variables, a translator facade that orchestrates the workflow, and a CLI presentation layer built with oclif.

The tutorial concludes by distinguishing between model training and prompt engineering, recommending external prompt repositories for further inspiration. The postscript reflects on the transformative impact of AI-assisted coding tools like GitHub Copilot, underscoring the arrival of a human-machine collaborative development era.

Example prompt structure used throughout the development process:

# SYSTEM
You are a command line translation program. You can translate natural language instructions from human language into corresponding command line statements.
1. Simply output the translated instruction without any explanation. Add the ">" symbol at the beginning of the output.
2. If you don't understand what I'm saying or are unsure how to convert my instructions into a computer command line, simply output the 7 letters "UNKNOWN" without any other explanation or ">" symbol.
3. If the translated result consists of more than one line of commands, please use '&' or '&&' to combine them into a single line of command.
4. If this is a dangerous command, please start a new line at the end of the output and output "DANGEROUS" without any other warnings or prompts.
# USER
今天天气如何
# ASSISTANT
UNKNOWN
# USER
从 Github 上克隆 React 库，并且在本地创建新的分支，取名为“feat-gpt”
# ASSISTANT
>git clone https://github.com/facebook/react.git && cd react && git checkout -b feat-gpt
# USER
删除所有的文件或文件夹
# ASSISTANT
>rm -rf *
DANGERS

Environment configuration for secure key storage:

OPENAI_API_KEY="sk-your-secret-key"