Useful context can include model structure, approved terminology, workspace rules, selected boards, operating definitions, and scoped internal guidance.
If the underlying model, policy, or workflow changes, update the agent context as part of the same release. Stale instructions are one of the fastest ways to reduce trust.
