GitHub's top design skill author reveals his loop: model-as-judge until 9.2/10
PrajwalTomar_ · x · 2026-11-16
The author behind GitHub's top trending design skill shared his AI design method—the loop matters more than the prompt:
- 10 real sites as references, not a paragraph
- 3 mockups on one page, pick one
- ban the AI look by name: no Inter, no purple gradients, no 3 equal cards
- design first, animations after
- a different zero-context model judges every round until it hits 9.2/10
Prajwal Tomar ran it on his burger brand: 6 rounds, 37 minutes, judge score went from 7.6 to 8.6. "Almost nobody is doing that last step," he notes.
More from coding & agent
- Scenario launches free AI school for creatives with 17 courses and 90 lessons — smtabatabaie · 2026-11-16
- Open-source AI Engineering course ships its own agent tutor via SKILL.md for Claude Code and Cursor — ghumare64 · 2026-11-16
- Embed Claude inside your artifact: each user's usage bills to their own plan — TawohAwa · 2026-11-16
- Dev runs 120B models on iOS by streaming MoE weights from flash in llama.cpp — Chance-Resource-4970 · 2026-11-16
- Claude quietly became a full workspace: a map of One Claude, Code, Design and Duet — alex_verem · 2026-11-16
- Claude Code adds experimental Code Mode: one JS script runs multiple tool calls in one step — Arindam_1729 · 2026-11-16