GenPark AI Agent Skill - Synthetic Text N-gram Diversity & Entropy Scorer

Computes Distinct-1/2/3 ratios and Shannon word entropy to detect generative model mode collapse and degenerative repetition loops.

Verified by GenPark AI and compatible with Model Context Protocol (MCP).

Architecture Diagram

graph TD
    A[Synthetic Generated Text Output] --> B[Token Normalizer & N-gram Slicer]
    B --> C[Distinct-1 Unigram Diversity Calculator]
    B --> D[Distinct-2 Bigram Diversity Calculator]
    B --> E[Shannon Information Entropy Engine]
    C --> F{Distinct-2 < 0.40 or Degenerate?}
    D --> F
    E --> F
    F -->|Yes| G[Flag MODE_COLLAPSE & Abort Stream]
    F -->|No| H[Approve High-Diversity Output]

Features

• Mode Collapse Tripwire: Detects degenerate token repeating loops (hello world hello world...).
• Zero External Dependencies: Pure Python standard library collections and math.