LilMGenius/paperthin

What it solves

Paperthin addresses the problem of "AI slop"—the tendency for AI agents to continuously add boilerplate, redundant files, and noise to a project rather than refining or removing it. It provides a set of low-level agentic design patterns (skills) that prioritize restraint and subtraction, ensuring that artifacts remain clean, maintainable, and free of "rot" (stale deltas and duplicated noise).

How it works

Paperthin is a collection of skills that can be installed globally across various AI agents (such as Claude Code, Cursor, Copilot, etc.). These skills are designed to be invoked either by the user or the model itself. Instead of adding more content, these skills focus on removing noise and refining existing work. For example, the re0 skill rewrites a drifted artifact into a clean version zero rather than patching it, and ssotize consolidates scattered facts into a single source of truth.

Who it’s for

It is for developers and users of AI agents who want to maintain high-quality, lean repositories and avoid the accumulation of AI-generated technical debt and redundancy.

Highlights

• Subtraction-first approach: Every skill is designed to remove noise, bloat, or redundancy rather than adding to it.
• Broad agent compatibility: Works across a wide range of agents including Claude Code, Codex, Cursor, and Grok-Build.
• Diverse skill set: Includes tools for rewriting (re0), verifying instructions (readchk), sizing compute effort (modelchk), adversarial planning (hate), and context-free auditing (shower).
• Iterative learning: Features a "coil" system (re0-loop, re0-memo) to ensure learning compounds across project iterations without bloating the codebase.