Is it intentional that the summary prompt doesn't include the user's prompt?

[seiya-koji]

buildSummaryPrompt()
builds the summary prompt from last_assistant_message only —
session.user_prompt is passed in by both providers but never referenced.
Is that intentional (token budget?), or just how it ended up?

The side effect: the observer fills in the <request> field ("what the user
asked for") without ever seeing the user's words, so it reconstructs them from
the assistant's reply — which tends to paraphrase intent into stronger mandates
(e.g. "investigate the bug" becomes "fix the bug"). Those summaries are then
injected into future sessions as context.