Most agent-to-agent messages and tool descriptions assume a human can ask for clarification. Downstream models cannot. This skill applies the discipline of ASD-STE100 (one meaning per word, active voice, simple tense, one instruction per sentence) to English that another agent or tool must parse automatically. It focuses on removing structural ambiguity without changing the underlying facts or scope.
What Sets It Apart
- Mode selection: a Strict mode for procedures, error messages and tool specs, and an STE-flavored mode for READMEs and explanatory prose. Modes change enforcement strength but keep the sentence-discipline.
- Deterministic structural linter: flags present-perfect/complex tenses, passive voice, multi-instruction sentences, oversized noun clusters (advisory), dangling conjunctions, semicolons, and a short list of soft phrasal verbs. It uses syntactic heuristics rather than full POS tagging to keep results predictable.
- Meaning-preserving rewrites: for each flagged sentence it rewrites the text without dropping any fact, condition, or scope qualifier. When a shorter phrasing would lose required precision, the tool keeps the longer wording and explicitly flags that trade-off.
- Minimal, pipeline-friendly output: by default it returns only the rewritten text (no preamble). Ask for a diff or explanation to get a before/after table and rule list. It intentionally does not reproduce ASD’s official word list and is not a certified replacement for the standard.
Who it's for and tradeoffs
Great fit if you build agent pipelines, tool descriptions, or system messages that downstream LLMs must parse with no human in the loop; it reduces misparses caused by ambiguous phrasing and complex sentence structure. Look elsewhere if you need certified ASD-STE100 word-by-word verification (this project applies the standard’s principles but not the official redistributable dictionary), if you expect creative or marketing prose, or if you require semantic equivalence proofs (the linter enforces form and preserves author-stated qualifiers but does not mathematically verify meaning preservation). The repo includes a CLI/skill integration path, example before/after cases, and deterministic lint rules tailored for agent-to-agent workflows.
