Make your LLM write like a Boeing manual, not a LinkedIn post

What it does

SimpleEnglish is a portable agent skill—one folder, zero dependencies—that plugs into any harness speaking the Agent Skills standard, including Claude Code, Cursor, Copilot, Codex, Gemini CLI, and roughly two dozen others. It overrides an LLM’s default prose with ASD-STE100 Simplified Technical English, a 53-rule controlled language born in 1983 for aerospace maintenance manuals. The result is flat, explicit text: active voice only, 20-word cap on instructions, no -ing forms, and conditions placed before commands so readers do not execute steps too late.

The interesting bit

The author treats the standard as a spec, not a vibe. Baseline agents hallucinated fake rule numbers and approved forbidden modals; this skill was built test-first against the primary Issue 9 text and benchmarked with a deterministic regex linter. It is essentially a negative image of every AI writing tell—run-on sentences, hedging, and telegraphic ambiguity all die as side effects.

Key highlights

• Measured 72.9% fewer STE violations per 100 words across six Claude models, with output tokens dropping on every one.
• A blind pairwise judge preferred the skill’s output in 38 of 48 head-to-head pairs.
• Ships adaptations for error messages, runbooks, incident reports, release notes, and system prompts—anywhere a reader cannot ask clarifying questions.
• Explicitly refuses marketing copy, blog voice, and brand writing; flatness is the feature, not a bug.
• Unofficial and unaffiliated with ASD; default mode is pragmatic, while strict mode approaches the standard more closely.

Caveats

• The skill does not produce STE-certified output, because ASD certifies no tool; word-level rulings still live in the official standard.
• A separate Pi cross-check showed the skill reduced violations on four additional models, yet three GPT-5.6 variants wrote slightly more words with it active.

Verdict

Worth a look if you generate technical docs, error messages, or runbooks and are tired of AI slop that hedges, rambles, and buries the verb. Skip it if you want personality, brand voice, or prose that sounds human at the expense of precision.