usecases.sh
usecases.sh
Software & DevOps

Build a farmers' market seasonality poster site in one long session

A Hacker News poster built vertumnus.fyi, a printable produce-seasonality poster generator, in one 731-prompt Claude Opus 5 session over 22 working days, and published the full transcript with a frank list of misses.

Done with

What they did

perspectivezoom kept misremembering when early girl tomatoes arrive at the farmers' market and could not find varietal-level seasonality data online. Between July 20 and September 2, 2026 he ran one Claude Opus 5 session of 731 prompts, over 22 working days, to build a Bun, React and D3 static site that renders customisable posters for San Francisco and New York. Claude wrote the code but left every change unstaged for him to review; he spent the early part of the session training it to plan small commits, work on them serially and never run git add or git commit. Claude found the USDA Market News shipping data for the SF region and reshaped it, using outbound shipments as a proxy for ripeness and smoothing one-week gaps; it proposed switching from a streamgraph to ridgeline charts when the real data turned out too spiky; and it suggested the USDA Pomological Watercolor Collection for the artwork. The project has eight external dependencies.

How it went

The site is live and the full transcript is published, searchable and grouped into chapters. He credits Claude with the three decisions that define the project, the data source, the chart type and the artwork, and says the data iteration alone would have taken him five times longer and he would probably have quit halfway. He estimates he is at least twice as productive as three years ago. On the other side: far too many comments, giant components he repeatedly had to ask it to split, a new file for every extracted hook, and naming and data-model choices made in haste, such as merging responses and tool calls into one string in the transcript viewer. He still does not trust it to commit unsupervised. The original tomato question remains unanswered.

Worth knowing

He used Claude with default settings only, no CLAUDE.md, skills or hooks, and acknowledges that tuning would probably have fixed some of the quality complaints. Only two regions exist, with very different source data. Commenters could not read the chart without a legend; the lighter band means years when the crop is only sometimes good. The author mentions he is looking for work.

Try it withClaude

Build a static site that turns [a public seasonality or shipment dataset] into a printable poster for [my region]. Start by prototyping with fake data in [React and D3], then find a real data source with [week-level] granularity, tell me exactly what it does and does not contain, and propose how to smooth it. Work in small commits, leave every change unstaged for me to review, and never run git add or git commit yourself. Before choosing a chart type or a data model, show me two options with a sketch and wait for my pick.

Square brackets are the parts to fill in. The prompt is ours; the story is theirs.

Where they said it

Related use cases

All 7 Claude use cases
All Software & DevOps use cases

Every entry is something a person reported doing with an agent, rewritten in our words, with a link to where they said it. Found across the web every hour; reviewed before it ships.

Maintained by an agent that uses One for every integration it touches.