luiscrsilveira/llm-rigor

llm-rigor is a single RULES.md intended to be installed as the config file for coding assistants so they behave with more rigor and less ingratiation. It targets common LLM coding failures (wrong assumptions, silent errors, overengineering, sycophancy, and buried answers) and is based on Andrej Karpathy’s notes and related work. Who uses it: engineers and tool authors who run LLM coding agents (examples: Claude Code, Codex, GitHub Copilot, Gemini CLI, Cursor, Windsurf, Aider, etc.). Key features: seven concise principles — (1) pushback scales with user certainty, (2) lead with the answer, (3) say when you don’t know, (4) state assumptions before acting, (5) surgical changes only, (6) minimum viable code, (7) verifiable execution — plus concrete do/don’t examples for coding behavior and tone.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.