Claude should proactively update CLAUDE.md and memory files during sessions, not just when asked

Currently, Claude only writes to CLAUDE.md and memory files when explicitly prompted to do so. For users working across multiple projects in longer sessions, this means context learned mid-session is lost if the IDE restarts, Windows reboots, or the session ends unexpectedly — because memory is only saved at the end if the user remembers to ask.

The expected behaviour would be: whenever something significant is learned or built (new file added, architectural decision made, deployment behaviour changed, gotcha discovered), Claude should update CLAUDE.md and/or write a memory entry immediately, then prompt the user to commit CLAUDE.md if it changed.

The memory infrastructure already supports this — the gap is that proactive writing isn't part of Claude's default behaviour. Most serious users working on real projects across multiple sessions would benefit from this being the default.