Hi, I'm Marcin Urbanski.
Ten years of graphic and web design, then development: WordPress, WooCommerce and Laravel for UK clients, checkout flows and performance work. Now I apply the same habit to agents. Reproduce, measure, fix the cause.
Opus, Sonnet and Haiku, four setups, three requests, 30 runs each. A written rule, installed as a skill as well, held against "skip the tests". Nothing I tried held when the user narrowed the scope themselves.
The process skills I run Claude Code with: test-first, verification, specs and independent review, each change checked with evals.
A native macOS time tracker, written by agents under my rules, tests and review.
Questions about agents, evals or a project? Write to me.