ai-ergonomics

From README

AI Ergonomics How comfortable is this codebase for an LLM to navigate, understand, and generate correct code in? Code that's ergonomic for AI is also better for humans — but the emphasis here is on what specifically helps or hinders LLM tools like Claude Code, Copilot, and Cursor. Context Window Friendliness LLMs work with a finite context window. Every file they read consumes tokens. Large files are expensive, and when a file exceeds the context budget, the LLM works with a partial view — leading to hallucinations and missed dependencies. What to check: Files over 500 lines deserve scrutiny. Over 1000 lines is a strong signal to split. Functions over 50 lines force the LLM to track too much state. It will lose track of variables, forget early conditions, or hallucinate return values. Deeply nested code (4+ levels) is hard for LLMs to reason about correctly — the same way it's hard for humans, but LLMs fail more silently. What to recommend: Extract focused modules.