<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> This repository tackles the proliferation of generic, AI-generated content ("vibe-coded" sites, prose, code) by identifying common detection cues through data mining. It targets engineers, designers, and creators aiming to distinguish and mitigate AI artifacts, fostering more authentic digital outputs. The project provides reproducible scripts and practical Claude skills for auditing and refining content.

How It Works

The project mines over 3.2 million Reddit posts across 47 subreddits for discussions on AI-generated content. Core analysis focuses on 3,033 comments from 125 canonical threads, tabulating specific "tells" (design features, writing cadences, code patterns) flagged by users. This data-driven approach, verified against real quotes, offers an objective ranking of AI detection cues. Findings are packaged into Claude skills for practical application and reproducible Python scripts for deeper analysis.

Quick Start & Requirements

Claude skills (unslop-ui, unslop-text, unslop-code) install by unzipping .skill files to ~/.claude/skills/ or via the claude.ai UI. Analysis scripts require pip install -r requirements.txt and Python 3 execution. No API keys or external authentication are needed. Official quick-start or demo links are not explicitly provided.

Highlighted Details

• Analysis based on 3.2M posts and 3,033 comments from 125 on-topic threads.
• Headline finding: AI sites are recognizable by sameness ("They all look the same" is the loudest complaint).
• Top visual tells: shadcn/Tailwind defaults and "AI purple" gradients.
• Reproducible Python pipeline for data harvesting, analysis, and chart generation.
• Companion studies and Claude skills for AI-written code and text.

Maintenance & Community

No specific details regarding project maintainers, community channels (e.g., Discord, Slack), sponsorships, or roadmaps are present in the provided README.

Licensing & Compatibility

Code is MIT licensed, permitting broad use, including commercial applications and closed-source linking. Harvested text data originates from public Reddit content and belongs to its original authors; refer to DATA_NOTE.md for specifics.

Limitations & Caveats

Analysis represents a proxy for vocal online opinion; relative rankings are more reliable than exact percentages. Keyword matching may introduce noise or misinterpret sarcasm. Small subreddits can skew results, and some terms were rejected as keyword artifacts. Newer "tasteful defaults" may simply replace older ones.