Ten files. Read them in order.

Entry point for the agent. Modes (audit vs fix), when to skip, the workflow, and the top-20 lethal items.

Density formula, severity tiers, genre adjustments, model fingerprints, contested-tell handling. Read this first.

45 rhetorical and structural patterns across 5 groups. The bulk of the AI-slop axis.

~150 word-level tells across 7 categories: LLM verbs, cliché metaphors, intensifiers, sycophancy, weasels, connectors, spike words.

~33 markdown and structural patterns: bold-first bullets, title clichés, repetition, whitespace artifacts.

35 patterns across 5 groups for the comprehension axis: density overload, telegraphic compression, audience assumptions, scannability, cognitive friction.

Eight readability formulas with audience targets and the three cold-reader density signals (acronyms, named entities, numerics).

Canonical dual-verdict output: AI-slop and Comprehension scores, top fixes, mechanical and qualitative violations, recommended action.

~50 published sources backing every pattern. Peer-reviewed linguistics, vendor methodology, practitioner literature, viral takedowns.

Python regex scanner. Density-aware scoring, genre flags, model fingerprinting. JSON / quick / human modes.