wtf is founder-coding-skills?

founder-coding-skills is a set of thirteen skill files for Claude Code, aimed specifically at founders who have never written code themselves and are relying entirely on an AI to act as their engineering team. It is not written for programmers who review the AI's diffs and catch its mistakes. It assumes the person running it has no way to check the code, so the AI has to be disciplined enough to be trusted on its own. The pack splits into two layers. The first layer is aimed at the founder and covers things like how the AI should interview you before starting work, how to keep humans and AI in their proper roles, how to estimate cost before launching multiple AI agents at once, and how the AI should report bad news honestly instead of only good news. The second layer disciplines how the AI actually writes code: slicing work into small visible pieces, requiring a failing test before a passing one, investigating bugs properly instead of guessing at fixes, and reviewing finished work on two separate questions, whether it was built correctly and whether it was the right thing to build in the first place. The author states plainly that this pack is designed and tested only for Claude Code running Anthropic's most capable models, the tier available through a Max subscription, and that it has not been verified on weaker models or other AI coding tools. The skills are written briefly on purpose, giving the AI goals and boundaries rather than detailed step by step instructions, based on the idea that heavy hand holding helps weaker models but can hold back stronger ones. The author includes a real account of a mistake: burning around five billion tokens of usage in three days before writing these rules, along with a case study describing what went wrong and how the new rules cut that cost down. Installation is a single git clone and copy command, after which the skills load automatically inside Claude Code without needing to be invoked by hand.