task -agent

A simple task agent built from scratch in Python as part of the Foundations of Agentic AI learning journey.

Overview

This project demonstrates the core architecture of an AI task agent without using any agent framework.

The agent follows the:

Perceive → Reason → Act → Observe

loop and decides which tool to use based on the requested task.

Features

Custom agent loop built from scratch

Dynamic tool selection

Multiple tools

Scratchpad for storing observations and results

Execution trace for each step

Graceful handling of unsupported tasks

Double stopping conditions to prevent infinite loops

Tools

Calculator

Handles mathematical expressions.

Example:

run_agent("Calculate: 12 * (5 + 3)")