The claimed reduction in security is addressed in a response here [...] essentially the SHA512 hashing step invalidates the proposed attack.

If they're going to reduce the size of the seed by 50%, I would like to see some audit attention about whether this choice makes the protocol insecure.

The audit doesn't give me any confidence that this is secure. Session characterizes the audit like this:

Session’s generation of Ed25519 keys using 128 bits of entropy was explicitly identified in Quarkslab’s audit of Session, and Session developers had similar discussions with the Quarkslab team. Ultimately, they classified this finding as “low” because although the approach was non-standard, there was no practical nor theoretical method found to exploit this non standard approach.

I don't believe this characterization of the audit. I think that if the auditor found an vulnerability, then later realized that the vulnerability was not really exploitable, the vulnerability would be removed from the final report.

Instead, the issue is still in the report, which tells me that Session and their auditor weren't able to come to agreement about whether the seed size reduction is a vulnerability. Instead, the auditor included Session's response in the report, neither agreeing or disagreeing with it. This tells me that they either don't agree with Session's position, or their auditors don't have enough familiarity with crypto to evaluate if Session is right. Either one is worrisome.

I also don't place much importance on the Low rating. These ratings are, to some extent, negotiable.