When the Agent Is Wrong: Surface Real Kubernetes Errors, Not Model Guesses

Kubernetes AI agents should prioritize preserving real error messages over providing explanations. This is because real error messages contain actionable evidence, such as Kubernetes API errors, stderr, and exit codes, which are crucial for incident resolution. Agents should summarize these messages, but not replace them with speculative explanations. When reporting errors, agents should use evidence-backed language to provide clear next steps for human operators.