ASD-STE100 Skill

Status check 2026-11-13 — active

Verdict: The latest commit on master is d5caf549, dated 2026-11-12 (PR #66, a linter fix for sentence ends at trailing emphasis markers). Last release: no releases (commits only); no tags · last commit: 2026-11-12 · licence: MIT. The official ASD-STE100 dictionary (about 900 words) is deliberately not included… Repo: https://github.com/danyuchn/asd-ste100-skill Changed since this note: The two weeks since the note brought a burst of fixes to the deterministic linter (scripts/ste-lint.py). It now lints the front-matter ‘description’ key (relevant for SKILL.md files) and supports MkDocs admonitions.

🚀 Description

danyuchn/asd-ste100-skill repurposes ASD-STE100 Simplified Technical English — the controlled language the aerospace industry built so maintenance instructions cannot be misread — for a new reader: an AI agent parsing another agent’s output, a tool description, an error message or an inter-agent instruction, with no human to ask “did you mean X or Y?”.

STE rules in short: one meaning per word, active voice, simple tenses, one instruction per sentence, short sentences, no dropped words.

🧩 How it works

  1. Picks a mode — Strict (procedures, error messages, tool descriptions) or STE-flavored (READMEs, PR descriptions; sentence discipline without the fixed vocabulary)
  2. Flags violations sentence by sentence: ambiguous words, present-perfect, unclear passive, multi-instruction sentences, noun clusters, dropped words, long sentences, phrasal verbs, nominalizations, semicolons, hedge stacks, marketing adjectives
  3. Rewrites each flagged sentence without dropping any fact, condition or scope qualifier; if a shorter phrasing loses precision it keeps the longer one and says so
  4. Outputs only the rewritten text (plus a Kept as-is: line); ask “show the diff” for a before/after table naming the rules

Includes a deterministic linter (scripts/ste-lint.py) for structural patterns: semicolons, a short list of phrasal verbs, nominalizations, marketing adjectives, passive voice, present-perfect, long sentences, synonym rotation. It does not check noun-cluster length, does not prove meaning was preserved, and never flags hedges. It also does not ship ASD’s official ~900-word dictionary (redistribution needs ASD’s written authority).

Limits worth remembering: it fixes the form of a text, not its substance; it is wrong for creative or marketing writing.

Reasoning for

Directly applicable to the text my agents read: skill descriptions, AGENTS.md, tool descriptions in MCP servers, error messages from my scripts. See Skill Authoring Best Practices and Context Engineering — clear, literal instructions are part of the same discipline, and Matt Pocock’s shared-glossary idea in Matt Pocock Skills attacks the same ambiguity from the vocabulary side.

Alternatives considered


Template: tool