Open-Source Claude Code Skill Rewrites Agent Instructions Using 1980s Aerospace Rules
A Claude Code skill ports aerospace Simplified Technical English rules to tame ambiguous agent-to-agent prose and tool descriptions.
Takeaways
- New Claude Code skill ports aerospace ASD-STE100 Simplified Technical English rules to rewrite ambiguous agent-facing text
- Targets tool descriptions, error messages, and inter-agent instructions where no human disambiguates
- Two modes: Strict for procedures and errors, STE-flavored for READMEs and explanatory prose
- Flags passive voice, hedge stacks, nominalizations, oversized noun clusters, and multi-instruction sentences
- Related STE skill measured 72.9% fewer writing violations and reduced output tokens across six Claude models
- MIT-licensed, instruction-only, no network calls; install at github.com/danyuchn/asd-ste100-skill
A new open-source Claude Code skill applies a 1980s aerospace writing standard to a current agent problem: language that one model generates and another model misreads. The project repository packages principles from ASD-STE100 Simplified Technical English as a prompt-time rewriter for tool descriptions, error messages, system prompts, and inter-agent instructions.
Claude Code skills are reusable instruction packages that guide the model through specialized tasks. This skill audits and rewrites machine-facing text before another agent consumes it. Ambiguous wording at these interfaces can change tool selection, argument construction, and retry behavior when no person is available to clarify the instruction.
From torque instructions to tool calls
ASD-STE100 was developed for aerospace and defense maintenance documentation, where vague wording can cause a technician to perform the wrong action. The standard combines 53 writing rules across nine sections with an approved dictionary of roughly 900 words. Each approved word has one meaning and one part of speech.
- Sentence structure: Keep sentences short and limit each instruction to one action.
- Voice: Use active voice, with narrow exceptions, and identify the actor responsible for an action.
- Procedures: Write commands directly instead of embedding them in descriptive prose.
- Vocabulary: Use one consistent term for each concept. For example, start is approved, while begin and commence are excluded.
The audit works sentence by sentence
The skill does not redistribute ASD’s official dictionary. It applies the standard’s structural principles and checks each sentence for problems that can make instructions harder to parse.
- Ambiguity: vague terms, inconsistent word choices, and unclear references