RT

Project · Creator and maintainer

Agents Kindergarten

A testable playground for agent roles, reusable skills, evaluation, and improvement loops.

Updated

Agents Kindergarten mark — a robot on a playground holding a laptop, smaller agents climbing the frame; tagline a playground of learning by doing.
Project mark. Role-based agents share reusable skills, run scored tasks and improve from their evaluation history.

Context

Agentic AI is easier to understand through working systems than isolated prompts.

Constraint

Framework examples often mix orchestration, behavior, and scoring, making it hard to see what changed after an evaluation.

Contribution

I created a compact Python playground that separates agents from reusable skills, registers both explicitly, evaluates role-shaped tasks, and supports incremental improvement.

Approach

Small base contracts, registries, evaluators, and improvement strategies are coordinated through a simulated virtual-company layer.

Outcome

New roles and skills can be added without rewriting orchestration, giving experiments a repeatable and inspectable structure.

Evidence

The public repository includes example agents, shared skills, evaluators, improvement strategies, tests, and an end-to-end example.

Artifacts