[mixedCase]

I've had success asking it to specifically spawn a subagent to evaluate each work iteration according to some criteria, then to keep iterating until the subagent is satisfied.

[endymi0n]

I’ve had great success replacing it with Kimi 2.6