2.1.0: no sentence cap on the reply

Removes the five-sentence cap on the chat reply. The rule is gone from SKILL.md, the output style, the system prompt, the self-check, the Stop hook, and the linter. Requested in #35 by arybczak.

Why. A rescore of the committed 2026-09-02 replies shows the cap was met in only 5 of 16 replies. The same replies had 18.9% of sentences over 25 words, against 5.7% for release 2.0.0, which had no formatting rule. The cap did not hold and it did not shorten sentences.

The reply register keeps every other rule: prose only, answer first, no em-dashes, no contractions, no openers or closers, define a concept term at first use.

Metric change. reader_check() no longer counts over-cap sentences in visible_total, because the skill no longer asks for five sentences. The published reply figure moves from 86% fewer visible defects (406 to 58) to 95% fewer (218 to 11), recomputed from the same raw files by python3 evals/check_numbers.py. The counted classes are now em-dashes, bold, headers, and bullets. Those counts come from the 2026-09-02 runs of 2.0.1, which still had the cap. No run exists yet for 2.1.0.

Update:

claude plugin marketplace update simple-english && claude plugin update simple-english

Full list in CHANGELOG.md.