Pick and configure guardrail checks in the agent editor

In the UI: the Guardrails capability settings on an agent or harness is now an editor instead of a raw form. Add a gallery preset to append its checks, or add a single check and set its rule, stage and what happens on a match. Choose whether a policy judge or moderation check is answered by the utility model or the organization’s decision model. The mode reads Active or Advisory.

For agents: llm_judge and mcp checks are now valid on the output stage. They run once on the finalized assistant message, with the same caps, advisory downgrade and fail-open behavior as the tool stages.

See Guardrails and #4459, #4542.