Agent Skill to Force Docs in ASD-STE100 Simplified Technical English

TL;DR Highlight

An STE-rules skill cut visible reply clutter from 218 to 11 across 16 Claude replies.

Who Should Read

Developers whose LLM-written READMEs, runbooks, and error messages come out wordy and vague

Core Mechanics

Aerospace manual rules for LLM writing
- ASD-STE100 is a controlled language aerospace has used since 1983.
- It is built so a tired mechanic cannot misread an instruction.
- The skill makes LLMs write with that discipline.
Works across many agents
- Any agent that reads the Agent Skills standard works, including Claude Code, Cursor, VS Code Copilot, and Gemini CLI.
- Install with one line: npx skills add AminBlg/SimpleEnglish.
No skill support? Paste a rule block
- Paste the rule block of prompts/system-prompt.md into your system prompt, AGENTS.md, or .cursorrules.
- A 60-token version exists for tight budgets.
Rules are a testable spec
- Instructions max 20 words, descriptions max 25.
- Put the condition before the command.
- No should/would/may/might, and one word keeps one meaning.

Community Reactions

Fewer visible defects in chat replies
- Across 16 replies to 8 questions, em-dashes, bold, headers, and bullets fell from 218 with no skill to 11 with 2.0.1 (95% fewer).
- Those runs used 2.0.1 with the five-sentence cap, and no run exists yet for 2.1.0 or later.
- A blind judge preferred 2.0.1 over 2.0.0 in 14 of 16 pairs.
Fewer linter violations in documents
- On 8 sqlpipe writing tasks, violations per 100 words fell from 4.09 with no skill to 0.91 with 2.0.1 (78% fewer).
- It is one run per cell and a run moves by about 0.5, so the README says read it as parity or better, not a ranking.
The README admits measurement limits
- Judges are Claude models, so family bias is possible.
- A 2026-09-02 audit showed the linter numbers measure rule obedience, not what a reader sees.

How to Apply

Try the one-line prompt first
- Add 'Rewrite this using ASD-STE100 simplified technical English' to your current doc prompt and compare.
- Compare the result with the skill-enabled output on the same input.
Install the skill with npx
- Run npx skills add AminBlg/SimpleEnglish in your project.
- Then say 'rewrite this with simple-english' on one README.
Paste the rule block where skills are missing
- In Cursor and similar tools, paste the rule block of prompts/system-prompt.md into AGENTS.md or .cursorrules.
- Use the 60-token version if tokens are tight.
Write doc rules as measurable limits
- Put checkable rules in your team prompt, such as instructions under 20 words and condition before command.
- Count sentence lengths in the output to verify.

Try It

copy & run

# Paste into a Claude or ChatGPT chat as is (simplified version of the skill's rules)

Rewrite the text below with these rules.
- Instructions max 20 words, descriptions max 25 words.
- Put the condition before the command.
- Use active voice and simple tenses.
- Do not use should, would, may, might.
- One concept, one word. Do not mix check, verify, confirm.
- Keep articles and the word 'that'.
- No decoration words like 'crucial' or 'robust'. State the fact.
- No bold lead-ins.

{{document to rewrite}}

Next: give the same document only the one line 'write in ASD-STE100' and compare the results.

Terminology

ASD-STE100A controlled English standard made for aircraft maintenance manuals. Think of it as a coding convention for English.

Agent SkillsA standard where an agent reads one folder and changes how it works. Like a plugin for agents.

LinterA tool that counts rule violations automatically. Here it counts violations in English sentences.

Output styleA Claude Code setting that changes reply style. It only works when installed via the plugin.