simple-english

Classify each passage as procedural or descriptive, then apply the STE rules: one instruction per sentence, 20-word limit for instructions, approved modals only (can, will, must), and condition-first phrasing. Run scripts/voice_lint.py on the draft to measure remaining violations per 100 words.

Strict mode enforces every ASD-STE100 rule including length caps, no contractions, and approved vocabulary for procedures and safety text. STE-flavored mode applies plain words and direct syntax to reference docs and PRs, where length and contraction findings require judgment rather than automatic rewrites.

Run python3 scripts/voice_lint.py path/to/draft.md from the skill directory, or pipe text via stdin. The script outputs word count, total violations, violations per 100 words, and em dash counts, with no external dependencies required.

No, the linter is heuristic and not a certified STE checker. It measures mechanical form markers like banned words and passive voice, but judgment rules such as correct technical nouns require a human. Final approval rests with the writer.

Avoid STE for marketing pages, brand writing, fiction, personal essays, and any section whose job is to persuade or carry a distinct voice, since STE strips voice on purpose. Use a natural-voice rewrite skill like humanizer for those sections instead.

Yes, error messages follow the pattern of stating what happened, the cause, then the fix as an imperative. Agent instructions are treated as procedures: one instruction per sentence, no "should" since models read it as optional, and conditions placed first.