danyuchn/asd-ste100-skill

This project packages ASD-STE100 Simplified Technical English rules as a Claude Code skill. It rewrites dense or ambiguous English that a machine or non-native reader must parse without asking a follow-up question: tool and function descriptions, error messages, system prompts, and messages between agents. The README frames an agent reading another agent's output as similar to a mechanic reading a maintenance instruction—no author available to clarify, so one meaning per word and one instruction per sentence matter.

What changes in practice

You paste text and ask Claude to "disambiguate this" or "apply STE100 to this". The skill picks one of two modes: strict for procedures, error messages and tool descriptions; STE-flavored for READMEs, PR descriptions and explanatory prose, which keeps the sentence discipline but drops the fixed-vocabulary lockdown. It flags violations sentence by sentence and rewrites them while, by its own description, keeping every fact, condition and scope qualifier. The default output is only the rewritten text, plus a one-line Kept as-is: note when something was deliberately left unsimplified. Ask for "show the diff" or "explain the changes" to get a before/after table naming the rule broken. The repository includes before/after examples and a rule summary with citations (examples, rule references).

What the linter actually checks

The bundled Python linter is deterministic and structural. According to the README, it checks semicolons, a short list of soft phrasal verbs (such as "spin up" and "reach out"), nominalizations, marketing adjectives, passive voice, present-perfect forms, long sentences, synonym rotation, and dangling conjunctions in supported list items. It has explicit gaps: it does not check noun-cluster length, does not recognize phrasal verbs outside its list ("take off the panel" passes), and never flags hedges or modality. The README also states the linter does not compare an original with a rewrite, verify that requirement strength stayed the same, or prove meaning was preserved, so a zero-violation result only means the configured structural checks found nothing. The project's own prose needs --baseline 36 to lint, because rule tables quote the patterns they forbid; those findings are examples, not defects.

Setup and limits

The quick path is the skills CLI from the project root (npx skills add danyuchn/asd-ste100-skill), which sends anonymous install telemetry unless DISABLE_TELEMETRY=1 is set (skills CLI); alternative instructions in the README clone the repository into ~/.claude/skills/. The project does not reproduce ASD's official ~900-word approved dictionary, because Issue 9 restricts redistribution; it applies the principle of the plainest available word instead, so it is not a certified STE checker (ASD-STE100). It is also not meant for creative writing, marketing copy, or other text where voice and nuance are the point, and it changes form rather than substance—a paragraph with nothing to say becomes short, clean and still empty. The code is MIT-licensed (license), and a Hebrew port adapts the idea, though ASD-STE100 defines English only (Hebrew port).

HotRepo's take

The repository was created on 2026-07-20. HotRepo recorded one monthly Python board observation: +3,556 stars this month and best rank #7 on 2026-10-30. The sample covers a single recorded date, so it does not show a trend in either direction.