Agent Skill Forces LLMs to Write Docs in Aerospace's Plain-English Standard

A new open-source Agent Skill makes LLM coding assistants write documentation in ASD-STE100 Simplified Technical English, the controlled language aerospace has used since 1983 to keep maintenance instructions unambiguous. The rules — 53 numbered directives across nine sections, including hard caps like no sentence over 20 words — are built as specs rather than vague guidance such as “write clearly,” which the author argues models ignore. A convenient side effect is the suppression of AI writing tics: in the project’s benchmark, the word “seamlessly” never once survived.

The skill installs via a one-folder, dependency-free package (MIT-licensed) and works across roughly 30 harnesses that support the Agent Skills standard, including Claude Code, Cursor, Copilot, Codex, and Gemini CLI. For tools without native skill support, a system-prompt version — down to a ~60-token variant — can be pasted into custom instructions or AGENTS.md files. It ships tuned adaptations for error messages, runbooks, incident reports, release notes, and even system prompts themselves, while deliberately refusing marketing and brand copy, where flat prose is a liability rather than a feature.

The author backs the claims with a reproducible benchmark: 6 Claude models across 8 tasks in two conditions (96 runs), scored by a deterministic regex linter. Results show a 72.9% drop in STE violations per 100 words with the skill loaded, shorter output on every model, and mean sentence length falling from 11.2 to 9.7 words. The project is careful to note it produces no official STE certification — ASD certifies no tools — and that it paraphrases the standard rather than reproducing copyrighted spec text.