AI agent skills for turning messy source data into reproducible datasets and clear data stories.
Give your AI coding assistant a source file or URL, and DataPressr's skills have it archive the original, build tidy CSVs with a rerunnable script, document the schema, sources and licence, and check the result. Then add analysis, charts or a short data story. The output is ordinary files in your own folder or Git repository; publishing to DataHub is optional.
Website and docs: datapressr.datahub.io
You need Node.js (for npx) and an AI coding assistant that supports skills. From the folder where you want your datasets:
npx skills add datasets/datapressrThen start your assistant in that folder and ask, for example:
Use DataPressr's archive and structure skills to turn this source into a dataset: [paste a URL or file path]. Then run the validate skill.
The setup guide covers the installer's options and each skill. The skills themselves are readable playbooks in skills/.
This chart is generated from the oil-prices dataset by a checked-in script, for the story WTI Went Negative. Brent Didn't. Browse all datasets and stories.
Work is tracked in Beads (bd); AGENTS.md holds the project conventions. Start a new agent session with "Read NEXT.md and follow it." The session prompt selects work from bd ready; the handoff protocol provides execution details and the migration audit records completed work and the Beads plan. Planning and evidence live in docs/; the published site is built from site/.
Rufus Pollock and Datopian, drawing on more than 20 years of data wrangling and open data experience.
The DataPressr skills, code and accompanying documentation are open source under the MIT licence. Datasets and archived source material retain their own licences, recorded in each dataset’s metadata.