Teaching AI Agents to Write Like Aircraft Mechanics

What it does

Repurposes the ASD-STE100 standard — the controlled-language grammar built so aircraft mechanics cannot misread maintenance instructions — as a Claude Code skill. It rewrites dense, ambiguous English into short, active-voice sentences with one instruction per sentence, targeting the specific case where one AI agent must parse another agent’s output without asking for clarification. It offers two modes: Strict for procedures and error messages, and a looser STE-flavored mode for explanatory prose like READMEs.

The interesting bit

The insight is that an LLM parsing a tool description is in the same bind as a non-native mechanic reading a torque spec: no back-channel, no author to query, and a single misreading can cascade. The skill doesn’t just simplify; it flags violations sentence-by-sentence — ambiguous word choice, passive voice with unclear actors, hedge stacks, semicolons, oversized noun clusters — and rewrites without dropping facts or scope qualifiers. When shortening would lose precision, it keeps the longer phrasing and flags the trade-off rather than silently mutilating the meaning.

Key highlights

• Two rewrite modes: Strict for tool descriptions and inter-agent instructions; STE-flavored for READMEs and PR descriptions.
• Mechanical rule checks with explicit citations — you can point to the exact word or punctuation that violates a structural rule.
• Respects ASD’s redistribution restrictions by not reproducing the official ~900-word approved dictionary, applying the underlying principle (plainest available word, used consistently) instead.
• Outputs rewritten text with no preamble; optionally shows a before/after table naming each broken rule.
• Explicitly preserves all facts, conditions, and scope qualifiers from the original text.

Caveats

• Fixes form, not substance: a vacuous paragraph comes out short, clean, and still empty.
• The official ASD dictionary rules are advisory rather than enforced, so this won’t produce certified STE-compliant documentation.
• Deliberately flat and literal by design — useless for creative writing or marketing copy where voice matters.

Verdict

Worth adding to your Claude Code workflow if you write system prompts, tool schemas, or error messages that other agents (or humans) must parse without follow-up. Skip it if you’re polishing prose where tone and ambiguity are features, not bugs.