DataPressr
An opinionated, skill-driven workflow for turning raw data finds into clean, published datasets — and eventually data stories.
DataPressr
Turn a raw data find into a clean, published dataset — the same way every time, whoever (or whichever model) does the work.
DataPressr is a set of opinionated skills (prescriptive playbooks) that carry a
dataset through its whole lifecycle. /init, /validate and /push already
covered the last mile — packaging and shipping. The newer skills cover everything
upstream: the judgement-heavy part that used to be done ad hoc in a chat window.
The workflow
| Stage | Skill | What happens |
|---|---|---|
| capture | capture | A URL or idea becomes a GitHub issue — near-zero friction, no judgement |
| archived | archive | The raw source is snapshotted with provenance (URL, date, licence) |
| structured | structure | Raw → tidy, typed CSV(s) + a real datapackage.json schema |
| — | init / validate / push | Scaffold, deterministic checks, publish to DataHub |
| enriched | enrich (planned) | Descriptive stats + first charts |
| story | story (planned) | A narrative write-up from one or more datasets |
The skills live in skills/
and install into any agent — npx skills add datasets/datapressr — not just Claude Code.
Honest status
- Solid: the
validatescript — a real test suite (npm test), 24 assertions. - One trial so far: the
structureplaybook has been run end-to-end exactly once (Project Drawdown), and on an easy case. Its cleanup-idiom module is unit-tested; the workflow itself needs a genuinely messy source to prove out — in progress. - Written, not yet exercised in anger:
capture,archive. - No data story exists yet. #9 is the first.
- Not designed:
enrich,story,monitor— deliberately waiting on 1–2 hand-written stories first.
Full decision history: docs/skills-vision.md.
Roadmap to v1
- Prove
structureon one genuinely messy source, from raw → typed dataset, with a committedbuild.ts. (#8) - Write 1–2 data stories by hand — plain Markdown + whatever charts are fastest. (#9)
- Decide a charting standard, driven by the friction from step 2.
- Write the
enrich+storyskills from what steps 2–3 taught. - Ship this site: landing → datasets → stories → docs. (#7)
Post-v1: monitor + unattended cloud execution (#6); skill evals for the prompt-only skills.
Pages
- What to review now — the short list of things that actually need eyes
- Datasets — what's been produced, with status and links
- Data story: The Keeling Curve — the first hand-written story
This site is site/ in datasets/datapressr, published with the Flowershow CLI (fl site --name datapressr --yes). Live at https://datapressr-rufuspollock.flowershow.me.
