Hands-on exercises for using coding agents in development research workflows. Each exercise is a self-contained folder with data, code, prompts, and (where relevant) agent skills. More exercises will be added over time as part of the DIME Analytics AI-enabled research training series.
These materials are designed for World Bank staff, research assistants, field coordinators, and others who work with Stata pipelines, survey data, and project deliverables—and want to see how agent skills change what a coding agent can do on realistic tasks.
Each top-level folder is one exercise. Open each folder to access the README with the steps for each exercise.
| Exercise | Exercise Instructions | What you practice |
|---|---|---|
| Main do-file | main-dofile/ |
In a very basic project setup, compare agent output with and without memory and skills when setting up a main do-file. |
| Agent skills (Lumora demo) | research-roles-skills/ |
Compare agent output with and without project-specific skills on a synthetic cash-transfer evaluation: deck updates, high-frequency checks, and reproducible analysis. |
Open the README inside each folder for setup steps, prompts, and checklists.
- A coding agent environment - for example, the Claude or the GitHub Copilot extension in VS Code or Cursor. See GitHub Copilot at the World Bank (link requires WB GitHub account membership) for World Bank setup instructions and how to request access from ITS.
- Main do-file exercise: Stata, if you want to run the generated master do-file locally; however, this is not required to complete the exercise.
- Lumora exercise: Stata 17 or newer to run
code/stata/MasterDoFile.do(optional for deck-only or HFC-only tasks).
- DIME Analytics — reproducible research tools and training
- DIME Wiki — coding guides, workflows, and best practices
- DIME AI — AI in development research
Questions or feedback: dimeanalytics@worldbank.org
Exercise data in the Lumora demo are synthetic. Welfare and operational indicators are for training only—not causal estimates of program impact.