What a plain language standard does to a coding agent

I've just posted a Medium post discussing my plain language plugin for Claude Code and Codex CLI. The plugin ships skills and an output style that push the model's prose towards plain language. On Opus 5 the effect was larger than I expected. Across six code reviews of the same buggy component, the styled replies ran 17 per cent shorter and used 32 per cent fewer sentences. They still found the same headline bugs. I then added a skill applying the same principles to code, which is not part of the standard itself. That turned into the more interesting experiment. I generated ninety implementations of one specification. With the code rules, the public function landed in the first quarter of the file in all ten runs, against five of ten with no rules. All ninety still passed the same 25 hidden tests. The plugin has since been updated to v0.6.1, and includes the plain language skill that applies to code. Repository: github.com/GaZmagik/iso-24495 - MIT Licensed