Most of the lab’s work rhymes: public data is stranded — removed, scanned, scattered — and then a real project has to be built around it so the result is reusable and accountable. We capture both halves as agent skills: portable, plain-Markdown instructions an AI assistant (Claude Code, Codex, Gemini, and others) follows to do the work the way we would.

There are two, and they compose. data-liberation gets the data out — orchestrating acquisition, cleaning, validation, and documentation from a government PDF, FOIA release, or scraped site into a tidy, documented dataset (it scaffolds from a project template). data-project architects the collaboration around that data — a right-sized repository, reproducible pipeline, documentation, governance, and an open-knowledge catalog, built only as far as the project needs.

Together they standardize how we acquire, clean, validate, document, and govern a dataset — the same method behind the At-Risk Federal Data Archive and our Center for Environmental Journalism collaboration.

They’re free to use and adapt. Read the practitioner guide, or bring a dataset to the help desk.