For example, lets say there are two people named, Bob, and Jim.
Bob wants to send Jim a long secret message, and he decides to use a OTP to encrypt his message. Both Bob and Jim know a secret password but it's only 8 character long, so they can't use this to encrypt the whole message. They are also unable to securely share a new key.
So Bob generates several random "child" keys, and then using their password as a seed for a CSPRNG he uses it to randomly select a series of characters from the "child" keys to form a new key, which he then uses to encrypt his message. Bob then sends the encrypted data with the "child" keys to Jim.
So, as long as Jim uses the same seed (their password) for the CSPRNG, the same sequence will be selected from the "child" keys, therefore giving him back the key that Bob used, which can then be used to decrypt the message.
Therefore, assuming an attacker has infinite computing power, he would soon realize that the message could be decrypted into a number of possible messages (just like the original OTP), and because he doesn't know what seed (password) was used for the CSPRNG, he'll never know for sure which message is the real one.
So, would this "adaptation" of the OTP actually work in the real world? What are the cryptographic weaknesses of using a OTP this way?