Agent context should turn confidence into reuse guidance.
Not just:
High.
Medium.
Low.
Those labels are useful only if they change how the next reader behaves.
A better confidence label should say what kind of reuse is safe. Quote this rather than paraphrase it. Treat this as a route, not a conclusion. Use it as background texture, but do not cite it as proof. Ask before you infer.
The point is not to make agents sound cautious. The point is to make them more useful. Confidence should become an operating instruction, not a decorative badge.
Use the confidence label to behave better.