← ALL PROJECTS
REF. 04 · Claude skills · spec generation

The Lovable Prompt

A spec-generation skill that turns a rough app idea into a structured, opinionated Lovable.dev build prompt — with defined data models, user flows, edge states, and acceptance tests. This site was built using it.

BUILT
May 2026
CATEGORY
Claude skills · spec generation
Claude.ai ProLovable.devMarkdown skill packages

1 pass
Rough idea → build spec
4 sections
Data, flows, edges, tests
This site
Built with it

PROBLEM

Lovable rewards specific, structured prompts and punishes vague ones. But the natural way to describe a new app idea is rough and conversational — by the time you've manually expanded it into data models, flows, edge states, and acceptance tests, you've spent an hour before the first prompt lands.

SOLUTION

The Lovable Prompt skill ingests a rough idea, runs structured elicitation to fill the gaps, then emits a full build spec organized exactly the way Lovable's docs recommend: data model first, user flows next, explicit edge states, and acceptance tests at the end. The output drops into Lovable as a single prompt and produces a coherent first build.

BENEFITS

  • Rough idea → ship-ready prompt in one pass — no manual spec writing
  • Opinionated structure (data → flows → edges → tests) matches Lovable's strongest prompting patterns
  • Acceptance tests make 'is this done?' answerable before the first build
  • This portfolio site was built end to end using it

CHALLENGES & WHAT I'D IMPROVE

Knowing when to stop eliciting — too few questions produces vague specs, too many turn it into an interview. Next iteration: a stakes-based questioning depth (more questions for production apps, fewer for throwaway prototypes) and a built-in iteration mode for refining a spec after the first build round.


← Back