v4.0 — Self-contained SKILL.md, morph validator, ReadSightPy, scenario unification

[MADEVAL]

v4.0 — Major Upgrade

Self-contained SKILL.md (~1200 lines)

• Embedded top-15 burned words, openers, transitions, hedging × 9 languages
• Embedded specificity ladder examples, tone key-markers, opener categories, conjunctions
• GitHub raw URLs for full data lists
• Pre-flight guard: min text size (100w), weighted threshold, RU-specific opener adjustment

New Scripts

• morph-check.ps1 — morphological validator (pymorphy3 for RU/UK, spaCy for DE/FR/ES/PT/IT/PL)
• readability-check.ps1 — ReadSightPy integration (Flesch, LIX, Fog)
• run-eval.ps1 — external EVAL.md runner (OpenAI/Anthropic API)
• clause-check.ps1 — rhythm rule validator
• run-benchmark.sh — Bash benchmark runner

Data

• shared/morph/ — whitelist + blacklist for RU
• tests/benchmark/readability-baseline.json

Coverage

• 15 scenarios unified to single format
• +12 AI test files (de/fr/es/pt/it/pl × email + landing/social)
• +12 examples (de/fr/es/pt/it/pl × blog + email/landing/social)
• +12 annotations entries

Infrastructure

• README badge 1.0→4.0
• validate.ps1: +4 scenarios, SKILL.md↔shared/ consistency check
• CI: validate.yml updated

Old v3.0 code preserved in legacy branch.