Lumora Agent Skills Training Demo

This repository is a synthetic World Bank-style coding-agent training project. It teaches how reusable agent skills improve output on realistic development research and operations tasks.

Participants who cannot use git can download lumora-agent-skills-training-demo.zip from the repository root and unzip it locally.

The mock study is the Lumora Family Support Grant Evaluation, a fictional cash transfer program with baseline, follow-up, payment, field visit, survey log, and back-check data.

What Is Included

• Synthetic raw CSV data in data/raw/
• Participant-facing Stata pipeline in code/stata/
• Pre-generated Stata tables and figures for the manager deck exercise in outputs/
• A baseline deck for participants to update:
  • decks/baseline_round1_deck.pptx
• A source library of agent skills in skills/
• Exercise-specific skill setup instructions in .agents/README.md
• Role-specific exercises in exercises/

Run The Stata Pipeline

Open Stata from the project root and run:

do "code/stata/MasterDoFile.do"

The Stata pipeline imports raw CSVs, writes intermediate .dta files, creates the processed household panel, exports tables, figures, and QA notes.

Expected outputs include:

• data/intermediate/*.dta
• data/processed/analysis_household_panel.csv
• outputs/tables/*.csv
• outputs/figures/targeting_funnel.png
• outputs/figures/payment_provider_delay.png
• outputs/figures/welfare_change.png
• outputs/figures/enumerator_quality_flags.png
• outputs/qa_reports/stata_pipeline.log

Participant Tasks

Each exercise uses the same before/after workflow:

1. clear .agents/skills/ and run the task with a general coding agent;
2. evaluate the output against the exercise checklist;
3. copy in only the role-specific skills;
4. run the same substantive task again;
5. compare what changed and what still needs human judgment.

See exercises/comparison_prompts.md for the shared scorecard.

Skill Setup

The source skills live in skills/. For the workshop, copy only the skills needed for the current exercise into .agents/skills/.

See .agents/README.md for role-specific copy commands.

You can run a baseline prompt first, then activate the relevant skill set and compare the difference.

Managers:

• update decks/baseline_round1_deck.pptx with follow-up outputs;
• inspect the Stata do-files first;
• produce an update log mapping slide claims to generated outputs.

Field coordinators:

• run high-frequency data quality checks;
• produce an actionable follow-up list before the next payment cycle.

Research assistants:

• clean and construct the household panel;
• validate IDs and merges;
• generate tables and figures reproducibly.

Important Caveat

All data are synthetic. Welfare changes are descriptive monitoring evidence, not causal estimates of program impact.