Skill Forces AI Docs Into 1983 Aviation English

SimpleEnglish forces LLM-written documentation through ASD-STE100, a 53-rule controlled-language standard aviation maintenance crews have used since 1983, per its GitHub page.

The stakes are practical. A technician or support engineer reading hedged AI output has to guess at conditions instead of following a direct command. SimpleEnglish's benchmarks show the fix works: testing across six Claude models cut STE rule violations 72.9% per 100 words.

The rules read like a style guide for people who can't afford to misread a sentence. Max 20 words per instruction, active voice only, simple tenses, no hedging modals, and conditions stated before the command they modify. AminBlg built it as an installable skill, npx skills add AminBlg/SimpleEnglish, and says it works across roughly 25 agent frameworks.

The project has 817 stars on GitHub and hit the Hacker News front page with 309 points and 113 comments.

The GitHub page doesn't say whether the 72.9% figure holds for longer-form docs or just short instructional text. It also doesn't break down which of the 53 rules gets violated most often.

Constraints beat prompting

Conditions before commands isn't a documentation trick. It's how you'd structure any instruction you wanted followed correctly the first time. If that holds, the next version of this skill won't stop at docs.