Emil Kowalski ships /learn-taste skill that turns your design corrections into rules
yuwen_lu_ · x · 2026-11-13
Frontend developer Emil Kowalski released a /learn-taste skill that fixes the pain of re-correcting coding agents on the same design mistakes every session.
How it works:
- Finds every correction in a session, including implicit ones: direct file edits, repeated nudges, reverted changes
- Generalizes a single fix into rules for all similar components (a shadow fix on one card covers every card, popover, and menu)
- Never invents a reason: if you didn't say why, it proposes one and asks you to confirm
- Writes rules into DESIGN.md, or suggests a lint rule when automated checking would be more reliable
Usage: run it at the end of any session where you corrected the agent more than once. Over time DESIGN.md becomes a record of your taste that every agent reads before touching the UI. The author stresses the reasoning matters most — knowing the why lets agents extend rules across the UI.
Install: npx skills@latest add emilkowalski/skills --skill learn-taste
More from coding & agent
- With AI help, a team finally migrated all their infra to K8s after ragequitting years ago — DanielLockyer · 2026-11-13
- Agents trust other agents by default — one user's skeptical-agent fix for a real security hole — andrey_kurenkov · 2026-11-13
- A dead-simple canary test for drifting agent sessions: make it address you by name — brandon_galang · 2026-11-13
- Firecrawl MCP v3 launches with 3,200+ tools, 60x rate limits and parallel subtasks — gaganghotra_ · 2026-11-13
- Software abstractions are dead: agents now rewrite whole stacks per workload — DimitrisPapail · 2026-11-13
- Voice agent replaces onboarding questionnaires: rethinking SaaS for agentic platforms — strickvl · 2026-11-13