◇ Could this help me?
Yes for the goal, only with the exposure understood and handled, and that qualification is doing real work here. Wanting to direct a coding agent from a phone is reasonable and the official feature covers the same need, so the idea is not the problem. What makes this different from an ordinary convenience project is what is being exposed: a remote interface to a coding agent is remote access to a machine that can read files and run commands, so a public link protected by a username and password is a login to that computer handed to anyone who obtains both. The creator's instruction to stop sessions when finished is the right instinct and deserves more weight than it is given, since it limits how long the exposure exists. General practice only in the narrow sense that remote access to a working machine should be reached over a private network or an authenticated tunnel rather than published, and the official feature should be retested before rolling your own.
I want to be able to direct a coding agent from my phone, and I want to understand the exposure before I build anything. First, tell me what the official remote feature currently supports and what its limitations are, since that is the lower-risk path. If I still need my own, do not start by writing it: explain exactly what a remote interface to a coding agent exposes, specifically what an attacker who obtained the link and credentials could do on that machine. Then propose the safest viable setup, preferring a private network or an authenticated tunnel over anything reachable from the public internet, and tell me how to limit what the agent can touch while it is reachable remotely and how to confirm a session is actually stopped.