AminBlg/SimpleEnglish

SimpleEnglish – an Agent Skill for Plain‑English LLM Output

What it does

• Provides a plug‑in (called a skill) that makes large language models rewrite their replies in the style of ASD‑STE100 Simplified Technical English – the controlled language used for aerospace manuals.
• Two output modes:
  1. Plain – removes most formatting, limits sentence length, bans contractions, bold, headers, etc.
  2. Strict – adds the full STE‑vocabulary restrictions for near‑certified compliance.
• Works with any agent that follows the open Agent Skills standard (Claude Code, Cursor, VS Code Copilot, OpenAI Codex, Gemini CLI, Goose, OpenCode, etc.).

How to install

# generic skill installation (no runtime hook)
npx skills add AminBlg/SimpleEnglish

• For full functionality (session hook & output style) install the agent‑specific plugin, e.g.:
  • Claude Code:

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

  • Codex:

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

• If your agent does not support the skill format, you can copy the system‑prompt block from prompts/system-prompt.md into your own prompt configuration.

Typical usage

• After the skill is active, simply ask the model to “rewrite this with simple‑english” or request any technical writing (README, runbook, error message, etc.).
• The output style is identified as simple-english:simple-english in the agent’s settings.

Benchmarks & evidence

• Evaluated on 9 Claude models, Claude‑sonnet‑4‑6, and GPT‑4.1‑mini.
• Reported 95 % fewer visible defects (em‑dashes, bold, headers, bullets) compared with the same model without the skill.
• STE‑linter violations dropped from 4.09 per 100 words (no skill) to 0.91 with version 2.0.1 – a 78 % reduction.
• Human judges preferred the skill‑enabled output in 14 of 16 blind pairs.

Project structure

• SKILL.md – the formal Agent Skills definition (≈1,700 tokens, 53 numbered STE rules).
• prompts/system-prompt.md – ready‑to‑paste system prompt for agents without skill support.
• evals/ – scripts and raw results used for the CI‑generated benchmarks.
• tools/ste-dictionary/ – a helper that can extract the official STE word list from the free PDF (the repo does not ship the dictionary itself, respecting the standard’s copyright).

License

• MIT. The repository only paraphrases the STE rules; it contains no copyrighted standard text.

Who might benefit

• Developers who want consistent, unambiguous documentation generated by LLMs.
• Teams maintaining internal tooling or APIs where mis‑interpretation can cause costly errors.
• Anyone using an LLM‑powered coding assistant and needing the output to be easily readable by non‑technical staff.