AminBlg/SimpleEnglish

What it is

SimpleEnglish is an Agent Skills package that changes how an LLM writes technical text. It applies the style of ASD-STE100 Simplified Technical English, the controlled language used in aerospace documentation since 1983, to make instructions hard to misread. The skill works in any agent that reads the Agent Skills standard, including Claude Code, Cursor, VS Code Copilot, OpenAI Codex, Gemini CLI, Goose, and OpenCode. The package is one folder with no dependencies and an MIT license.

It has two registers. Plain is the default and applies prose rules: answer in the first sentence, no headers, bullets, bold, or em-dashes in chat replies, no contractions or opening/closing filler, and define concept terms in a few words. Documents follow additional rules: at most 20 words per instruction and 25 per description, condition before command, simple tenses and active voice, no should/would/may/might, one word with one meaning across the document, and no decoration that hides the fact. Strict mode adds vocabulary discipline from the STE dictionary when you name STE, ASD-STE100, or compliance.

Install and use

Any supported agent can install the skill with the skills CLI:

npx skills add AminBlg/SimpleEnglish

This installs only the skill. It does not install the Claude Code session hook or output style. For those, use the Claude Code plugin:

claude plugin marketplace add AminBlg/SimpleEnglish && claude plugin install simple-english@simple-english

The output style is named simple-english:simple-english. Select it under Output style in /config, or add {"outputStyle": "simple-english:simple-english"} to ~/.claude/settings.json. Codex users can install the plugin and hook with:

codex plugin marketplace add AminBlg/SimpleEnglish
codex plugin add simple-english@simple-english

Codex asks you to trust the hook before its first run. The hooks need Node.js. If your agent does not support skills, paste the rule block from prompts/system-prompt.md into your system prompt, AGENTS.md, or .cursorrules; the page includes a 60-token version. After installation, ask for technical writing or say "rewrite this with simple-english".

Benchmarks and limits

The project ships evaluation results and a linter. In its own benchmark on 8 chat questions with Claude models, version 2.0.1 cut visible reply defects by 95% compared with no skill, counted as em-dashes, bold, headers, and bullets across 16 replies. The counts come from 2026-09-02 runs of 2.0.1, which still had a five-sentence reply cap that 2.1.0 removed; no run exists for 2.1.0 or later. On 8 sqlpipe writing tasks, STE linter violations per 100 words fell from 4.09 with no skill to 0.91 with 2.0.1. The README notes that a 2026-09-02 audit showed the earlier 81.3% linter-improvement figure measures rule obedience, not what a reader sees. Judges are Claude models on Claude text, so family bias is possible, and the authors describe the document results as parity or better rather than a ranking.

Strict mode gets close to STE but does not produce STE-certified output; ASD certifies no tool. Word-level rulings are in the official standard, a free download from ASD. The repository ships a dictionary extractor under tools/ste-dictionary/ but no dictionary content, because the standard forbids reproduction without written authority. The project is unofficial and not affiliated with or endorsed by ASD or STEMG.

HotRepo's take

The repository draws on a clear engineering need: LLM replies are often padded with formatting and vague claims, while runbooks and error messages need short, testable instructions. The rule set maps directly onto those failure modes, and the project publishes raw evaluation files plus a CI check that recomputes headline numbers, which is more than most prompt packages offer. The HN thread (363 points, 121 comments on 2026-07-30) shows the main debate: one commenter argues a plain "rewrite using ASD-STE100" prompt is good enough dan_sbl, while others note that pretraining does not retain exact rule details Ferret7446. Another commenter reports using similar STE prompting to remove padding from LLM writing baq. The daily Python snapshot on 2026-11-03 recorded 4,166 stars and 50 stars gained, placing it at position 17 on that day. The benchmark caveats and the missing run for 2.1.0 are worth keeping in view when comparing versions.