A URL Is One Piece of Data

An operator ran co auth google on a headless server. There was no browser to finish the handoff, so the terminal URL was the only path forward. They copied what looked like the URL, pasted it elsewhere, and Google rejected it. Rich had wrapped the 786-character value at the terminal edge, turning one URL into several physical lines. When a spinner followed, terminal control sequences made the copied text even harder to repair.

The tempting answer was to improve the instructions around the link. That did not change the actual interface: on a remote box, stdout is not decoration. It is the transport between the CLI and the browser on another machine. A value that needs hand-editing before use is a broken value.

The complication was that the friendly output still matters. The command should explain why a browser is opening, and a human should be able to spot the link. Dropping Rich for the entire flow would solve wrapping by discarding the rest of the presentation.

The turn was smaller. The label remains styled, while the URL gets its own raw line with Rich soft wrapping enabled and markup and syntax highlighting disabled. The same path now serves Google and Microsoft, so the next long OAuth query cannot rediscover the bug under a different provider name. A regression test renders a deliberately long URL through a 24-column console and proves the output still contains exactly one URL line.

The lesson is that terminal output can be both prose and data, but each part needs its own rendering contract. Labels may wrap. Tokens, URLs, IDs, and shell commands must survive copying byte for byte.