I made Claude Code think about the design before it writes code

A developer released system-design-first, an open-source (MIT) skill for Claude Code that also works with Codex, Cursor and other coding agents, which forces the agent to do rough load math and write a short plan with trade-offs and open questions before writing code for larger features while keeping small fixes small. In side-by-side tests of the same tasks with and without the skill, the developer reports plans included real numbers and open questions instead of guesses, and a typo fix stayed a typo fix.

Coding agents build what the ticket says and miss what it doesn't say. You ask for a CSV export, it builds the whole file in memory, and it falls over on the biggest customer. Or you ask for a simple filter and get a plan with Kafka. So I made a skill for Claude Code also works in Codex, Cursor and others . Small fixes stay small. For bigger features, Claude first does some rough math on load, reads how the code works now, and writes a short plan with trade-offs and questions for you. Then it codes. If you ask, it also draws the architecture in the terminal. I tested it by running the same tasks with and without the skill. The plans got real numbers and open questions instead of guesses. A typo fix stayed a typo fix. npx skills add Baukaalm/system-design-first Free and open source MIT : https://github.com/Baukaalm/system-design-first https://github.com/Baukaalm/system-design-first