What is codesplain?

Codesplain is a Claude Code skill built for people who direct an AI to build software for them but cannot read the code it produces. It solves a specific trust problem: if you cannot read code, you have no way to check whether the AI actually fixed a bug, quietly added a new dependency, or changed how your data is stored. Codesplain gives you a second file you can actually read and check against what the AI tells you. That file is called CODESPLAIN.md and lives in your project. It has three parts. A map section describes five to ten recognizable areas of the project in plain language, such as what a section does, why it exists, and which files it touches, without ever listing every file one by one. A decision log records every meaningful choice made while building, written as three to five plain sentences covering what was decided, why, what the alternative was, and what might cause problems later. A glossary defines every technical term used, once each. You install it by copying the skill folder into your Claude Code skills directory. Running a command starts a new map on an existing project by reading the code, or lays down an empty starting point on a brand new one. You can also ask plain language questions about the project, such as why something is slow, and get an answer grounded in the map. Once the file exists, the skill keeps it updated automatically as you keep building. Small additions like a new library get logged with a short note in chat, while anything that costs money, locks you into a choice, or risks losing data stops and asks for your approval first. Minor changes like typo fixes or renames are never logged, so the file stays short enough to actually read. It is aimed specifically at people building real software through AI without a coding background, and is released under the MIT license.