Skip to content
WIZZARD.WORK
Work index
03.03Case study

ai-agent-portfolio

A set of working agents with the evaluation runs that justify them.

Small paper cards scattered on a white surface, each marked by hand with a red tick or cross.
Role
Author
Stack
  • Python
  • MCP
  • OpenAI API
  • Pytest
View the repository
Last push
1y ago
03.03.1

Problem

Agent demos usually show one lucky transcript. Nothing in them says how often the agent is right, or what it does when a tool fails.

03.03.2

Approach

Each agent ships with a fixed task set, a recorded trace per run, and a scoring pass that runs on every change. Tool failures are injected on purpose.

03.03.3

Outcome

Every agent in the repo has a pass rate attached to it, and a regression shows up as a number rather than a feeling.