Show HN: Microagents: Agents capable of self-editing their prompts / Python code

microagents: Modular Agents Capable of Self-Editing Their Prompts and Python code

Manufactured Agent Prompts (From Above Demo)

CalculateAddition Agent

You are an adept arithmetic solver with focus on performing addition. Utilize this Python function to calculate the sum of two numbers:

``python
def calculate_addition(num1, num2):
    return num1 + num2

# Example usage:
print(calculate_addition(5, 9))
``

GetPopulationOfCountry Agent

You are a skilled data extractor specializing in population statistics. Retrieve the population of a given country using the provided Python code:

``python
import requests

def get_population(country):
    url = f"https://restcountries.com/v3.1/name/{country}"
    response = requests.get(url).json()
    population = response[0]['population']
    print(f"The population of {country} is {population}.")

# Example usage:
get_population("CountryName")
``

How does it work?

This experiment checks out self-evolving representatives that immediately produce and enhance themselves. No particular representative style or triggering is needed from the user. Merely present a concern, and the system starts and develops representatives customized to offer responses. The procedure begins with a user inquiry, triggering a standard “bootstrap” representative, which does not perform Python code however strategies and delegates to specialized representatives efficient in running Python for more comprehensive functions. An Agent Manager manages them, picking or producing representatives through vector resemblance for particular jobs. Representatives have progressing system triggers that enhance through knowing. For coding jobs, representatives consist of Python in triggers, improving their technique through an “development action” if not successful. Upon finishing a job, a representative’s status updates, and the bootstrap representative examines the outcome, engaging other representatives for more actions in bigger procedures.

Existing Challenges and Potential Improvements

1. Course Optimization: The system often stops working to efficiently dispose of non-functional representatives.

2. Efficiency and Parallelization: Currently, parallel processing is not executed. Allowing the screening of several timely advancements all at once might considerably improve efficiency.

3. Method for Prompt Evolution: The method to trigger development is rather fundamental at the minute. Establishing an approach to measure the success ratio would improve this technique.

4. Consistent Agent Prompts: There is considerable capacity in incorporating consistent representative triggers with vector databases. In addition, sharing effective representatives throughout numerous runtime environments might enhance general effectiveness.

5. Hierarchical Agent Structure: Most demands are currently processed straight by a representative designated by the bootstrap representative. Carrying out a more elaborate hierarchical structure for handling demands might result in significant enhancements.

6. Context Size Limitation: Not yet thought about.