Agent context should make safer reuse feel practical.
Safety cannot mean the context becomes unusable.
It should be clear enough to follow.
What can be reused?
What should stay tentative?
What should be asked instead of assumed?
Good agent context turns those limits into a working path.