Failure modes and guardrails

Applying safeguards that keep AI agents from looping or drifting off-task.