This repository includes a collection of "tasks", each represented by a markdown file with a prompt which is meant to be used as input for an AI coding agent (e.g., gemini cli or claude code). The 'generated' folder has dated artifacts generated by the agent for each task.

What's this for? I'm using this repo to explore how coding agents perform for the kinds of data wrangling tasks we handle in the DREAM Lab. I want to understand differences between agents and for the same agent over time.

Here's an example of how to run a task. (This is very dangerous because we're running the agents without any guardrails. It's best to run this on an isolated VM):

mkdir tmp
cd tmp 

# run the task
gemini --yolo < ../task/covid-morans.md
# OR claude -p --dangerously-skip-permissions

# wait for the task to complete ... 

# save the generated outputs
cd ..
mv tmp generated/covid-morans/gemini/YYMMDD