Type the instructions you would give an AI agent, one step per line. The graph builds as you type, and tells you which steps act with no human approval before them. See the compiled examples.
Try it: add the line Wait for my approval above the first orange step, and watch the warning clear. The examples are illustrations, not any company's real procedures.
This page reads keywords, line by line, in your browser. Nothing is sent anywhere. It is a quick draft, not an audit: it can miss a risky step that is worded unusually, and it assumes the lines run in order. For real instruction files, compile them with a coding agent and the validator in the repository, which makes every step quote the line it came from.