Merge main v4.0 → legacy: self-contained SKILL.md, consistency fixes, human text descriptions

.github/workflows/validate.yml

@@ -10,3 +10,14 @@ jobs:
         run: pwsh scripts/validate.ps1 -Verbose
       - name: Run Bash validator
         run: bash scripts/validate.sh --verbose
+      - name: Check SKILL.md self-containment
+        shell: pwsh
+        run: |
+          $content = Get-Content SKILL.md -Raw
+          if ($content -match "raw.githubusercontent.com") { Write-Host "PASS: GitHub raw URLs present" } else { Write-Host "WARN: Missing GitHub raw URLs" }
+          if ($content -match "Top burned words") { Write-Host "PASS: Embedded burned words" } else { Write-Host "WARN: Missing embedded burned words" }
+      - name: List new scripts
+        shell: pwsh
+        run: |
+          $scripts = @("morph-check.ps1","readability-check.ps1","run-eval.ps1","clause-check.ps1","run-benchmark.sh")
+          foreach ($s in $scripts) { if (Test-Path "scripts/$s") { Write-Host "OK: $s" } else { Write-Host "MISSING: $s" } }

EVAL.md

@@ -1,6 +1,6 @@
 # EVAL - Quality Evaluation Framework for HUMAN-AI
 
-> **NEW in v3.0:** ZeroGPT external validator integration. Run `scripts/run-benchmark.ps1` with a ZeroGPT API key to get independent AI detection scores.
+> **ZeroGPT external validator integration.** Run `scripts/run-benchmark.ps1` with a ZeroGPT API key to get independent AI detection scores. See also `scripts/readability-check.ps1` and `scripts/morph-check.ps1` for multi-layer validation.
 
 > Load this prompt alongside SKILL.md output to score the humanization result.
 > The evaluator is a separate prompt to avoid self-assessment bias.

KNOWN_LIMITATIONS.md

@@ -8,7 +8,7 @@ LLMs cannot reliably count exact words. The rhythm system uses clause-count and
 
 ## Language Depth
 
-English, Russian, Ukrainian, and German have the deepest coverage with native-level tone markers and cultural notes. French, Spanish, Portuguese, Italian, and Polish have been enhanced in v4 but may still have gaps compared to the primary four languages. Community contributions from native speakers are welcome.
+English, Russian, Ukrainian, and German have the deepest coverage with native-level tone markers and cultural notes. French, Spanish, Portuguese, Italian, and Polish may still have gaps compared to the primary four languages. Community contributions from native speakers are welcome.
 
 ## Edge Cases
 
@@ -23,7 +23,7 @@ English, Russian, Ukrainian, and German have the deepest coverage with native-le
 Results vary significantly between models:
 - GPT-4, Claude 3.5+, Gemini 2.0: recommended, consistent results
 - Claude 3, GPT-4o-mini, DeepSeek V3: adequate, may need parameter tuning
-- Smaller models (<13B parameters): may struggle with the full 6-stage pipeline, produce inconsistent self-evaluation scores, or miss subtle AI markers
+- Smaller models (<13B parameters): may struggle with the full 5-stage pipeline, produce inconsistent self-evaluation scores, or miss subtle AI markers
 
 The self-evaluation score (Stage 5.6) is generated by the same LLM that did the rewriting. It is a self-assessment, not an independent verification. Always review output before publishing.