[Stronz]

I built this after repeatedly running into the same issue in longer task-oriented conversations: the model gradually drifts into explanation mode and becomes verbose, speculative, and initiative-heavy.

The idea is to initialize the conversation in a stable execution mode so responses stay structured and operational over longer threads.

The README shows the same prompt in two conversations: – default model behavior – the same prompt with the snapshot applied

Curious if others here have run into the same drift in longer conversations.