Skip to content

USAGE.md

Latest commit

 

History

History
89 lines (65 loc) · 2.42 KB

USAGE.md

File metadata and controls

89 lines (65 loc) · 2.42 KB
Raw

Usage Guide

This guide provides examples of how to use the prompts in this repository.

Using Prompts with an Agent

Most prompts are designed to be used with an LLM agent. You can load the prompt YAML file, parse it, and feed the messages to the LLM API (e.g., OpenAI, Anthropic).

Prompt Execution Lifecycle

sequenceDiagram
    actor Developer
    participant Agent Script
    participant YAML Prompt
    participant LLM API

    Developer->>Agent Script: run_prompt(prompt_file, variables)
    Agent Script->>YAML Prompt: Load .prompt.yaml
    YAML Prompt-->>Agent Script: Parsed Prompt Object
    Agent Script->>Agent Script: Inject Variables (e.g., {{name}})
    Agent Script->>LLM API: Send formatted messages
    LLM API-->>Agent Script: Return Response
    Agent Script-->>Developer: Output Result
Loading

Python Example

import yaml
import openai

def load_prompt(filepath):
    with open(filepath, 'r') as f:
        return yaml.safe_load(f)

def run_prompt(prompt_data, variables):
    messages = prompt_data['messages']
    # Replace placeholders
    for msg in messages:
        for key, value in variables.items():
            msg['content'] = msg['content'].replace(f"{{{{{key}}}}}", str(value))

    response = openai.chat.completions.create(
        model=prompt_data['model'],
        messages=messages,
        temperature=prompt_data['modelParameters'].get('temperature', 0.7)
    )
    return response.choices[0].message.content

# Example usage
prompt = load_prompt('prompts/technical/software_engineering/tasks/01_code_review.prompt.yaml')
result = run_prompt(prompt, {"code": "def foo(): pass"})
print(result)

Using Prompts Manually

You can also copy the content of the messages field and paste it into ChatGPT or Claude. Remember to replace any {{variable}} placeholders with your actual data.

Running Tests

To verify that a prompt behaves as expected, you can run the defined tests using a test runner (not included in this repo yet, but testData provides the cases).

Each prompt file may contain testData which lists input variables and expected output.

testData:
  - input: |
      project_name: MyProject
    expected: |
      Repository Overview

Search Prompts

You can search for prompts using the included script:

promptops search "code review"

Validation

Before submitting a new prompt, run the validation suite:

./scripts/validate_prompts.sh