Agent Skill to Force Docs in ASD-STE100 Simplified Technical English
Engineers are experimenting with ASD‑STE100, a controlled “Simplified Technical English” standard from aerospace manuals, as a way to curb verbose, jargon-heavy prose produced by modern large language models. Some see value in packaging this style guide as an “agent skill” or default output mode to make documentation clearer for non-native speakers and distracted readers, while critics argue a single prompt line or external linters can achieve similar results and that many such skills are low-effort cruft. The broader theme is frustration with increasingly wordy AI outputs and exploration of style constraints — from STE to the Economist guide and Orwell’s rules — as practical tools to make AI-generated text more concise, precise, and readable.
Perceived benefits of ASD‑STE100 for LLM output
- Many commenters like using Simplified Technical English (STE) to counter “AI slop”: verbose, over‑ornate, pseudo‑smart prose.
- Reported effects: shorter sentences, clearer instructions, less marketing tone, easier docs and emails, more tolerable output for distracted readers.
- Some use STE (or similar constraints) as a global instruction for agents and say it improves readability, though not “earth‑shattering.”
- Example of why STE exists: turning dense warnings into short, direct safety instructions, especially for non‑native speakers.
Skepticism about the dedicated “agent skill”
- Several argue a one‑line system prompt (“use ASD‑STE100”) works as well as the full skill, calling the repo “cruft” or low‑effort.
- Others reply that skills are useful as shareable, versioned bundles: instructions, style rules, troubleshooting notes, and helper scripts.
- Some describe an iterative workflow: watch the agent, capture mistakes as new skill rules, then refine those over time.
Prompting vs tooling and enforcement
- Concern that models drift from any style constraint over long interactions.
- Suggestions: use linters (e.g., Vale) and pre‑commit or post‑tool hooks to enforce style on docs and comments.
- One link claims Orwell’s six writing rules as a prompt sometimes outperform STE prompts for reducing “slop” with fewer tokens.
Model style problems and comparisons
- Multiple complaints about recent models (especially newer Claude/Opus variants) producing long, meta‑commentary‑laden, sometimes unreadable text.
- Some find newer GPT models respond more directly when asked for blunt or plain language.
- A few users note that “personality” instructions often just add more verbose sections rather than true concision.
Debate over STE itself and broader trend
- Discussion of specific STE examples and whether they are actually unambiguous; some say STE alone cannot fix ambiguity without domain context.
- Question raised whether constraining language could impair reasoning; some argue style constraints should ideally be a post‑processing layer.
- Several note a sudden spike in STE‑related posts, traced to a viral tweet; opinions split between seeing it as a useful discovery vs a passing AI‑productivity fad.