How designers lock repetitive work into skills with Claude Code
AIMade by AISummarised by AI (Claude) from one piece of material it read (a talk or an explainer).How it's made
The way the product design team at a recruiting platform has worked Claude Code into its day-to-day work comes down to four patterns. First, they paste in a template prompt from the in-house design system and spin up a local prototype from a single line of natural language such as 'make an event payment page', and the repetitive task of adding templates is registered as an add-template skill so that it runs in the right order even if the person in charge forgets the procedure. Second, in a pipeline of hand sketch → 3D conversion with Nano Banana → detailed edits in natural language → character sheet → motion generation with Kling, the character sheet serves as an anchor that prevents shape distortion from arbitrary generation, such as of the back view. Third, they control hallucination by adding a rule to the planning-document automation skill to mark anything it doesn't know precisely as 'needs checking'; fourth, giving a Figma link rather than a screenshot lets the model read even spacing, color and font values and produces far more precise results, so it is better to feed in the original structured data than a processed summary. Results such as a one-month development period and a week or less of late nights are self-reported, so they are noted for reference only.
Key points
- Register repetitive work as skills plus workflows to enforce the procedure
- Lock in consistency across repeated generations with reference sheets such as character sheets
- Control hallucination by writing a 'mark it as needs checking if unsure' rule into the prompt
- Accuracy rises when you give original structured data, like a Figma link instead of a screenshot
This is a note: a summary an AI (Claude) wrote of one piece of material it read (a talk or an explainer). It has no source link; events checked against a source are kept as records.