v2.0.0: plain English by default

Simple English 2.0.0 changes the default. The skill now writes plain English for a smart reader outside the field, with the discipline of ASD-STE100 underneath. Strict mode keeps full STE vocabulary compliance for the document when you name STE, ASD-STE100, or compliance.

What changed

• Default mode is Plain: the structural rules, common words over jargon, a short definition for concept terms at first use, "you" as the reader, the point first. Procedures keep their headed, numbered, imperative shape.
• Replies get a register of their own: answer first, five sentences or fewer, no filler openers or closers, no slop words.
• Strict mode material (dictionary-only rules, part-of-speech rulings, verb rulings, recurring errors) moved to references/strict-vocabulary.md and loads on demand.
• A 69-term slop lexicon (evals/slop.tsv, built from 122 published ban lists) feeds the linter alongside the measured core list.
• The Claude Code plugin gains two advisory hooks: a Markdown lint after Write and Edit, and a reply-register check on Stop. Neither blocks.
• The Codex plugin now loads its session hook (the manifest names hooks/hooks.json) and ships .agents/plugins/marketplace.json, so the documented install works.
• The session hook sends the condensed prompt instead of the full skill, under the 10,000-character cap Claude Code applies to hook output.
• The linter counts em-dashes as a violation (from #19), and the trailing-condition scan is linear (from #12).
• README rewritten in the standard-readme shape. A gpt-4.1-mini run joins the results.

Measured

• 2.0.0 against 1.3.0, same linter: documents within one violation per run on claude-sonnet-4-6 and gpt-4.1-mini; replies 2.94 to 1.68 violations per 100 words with zero slop hits. Full table and raw files: evals/results/pivot-2026-09-01/RESULTS.md.
• The 74.6% headline in the README was measured on 1.3.0 and says so.

Thanks to @brunosaraiva-dvt (#12), @alexxozo (#14), @NixFrog (#19), @omegagussan (#20), and @KatSuricata (#13).