Skip to content

LLM authoring

An 'edit with ChatGPT/Claude' button that produces a config file to paste back.

Decision: ⚪ idea

Context

The editor offers a button that hands the creator a ready prompt - a skill file documenting the full JSON schema: questions, weights, orientations, result modules.

The creator runs it on their own LLM account, works the design through with the agent, and returns with a JSON file they paste into the editor and refine by hand.

Opportunity

  • Drops the entry barrier from "design a scoring model" to "describe an idea".
  • Zero inference cost to us - no model integration, no per-token bill, no vendor lock-in. The skill is a static, versioned artefact.
  • Hosted generation stays open as a later step, once volume justifies paying for it.
  • Lowers the cost of authoring for everyone, not only community creators.

Risk

  • The copy-paste handoff is clunky and is a real drop-off point.
  • Output quality depends on the creator's model and plan, and we control neither.
  • Schema drift silently breaks pasted files - the skill has to be versioned together with the schema.
  • Cheap generation means cheap bulk submissions, pushing load onto moderation.