agent-eval

name	agent-eval
description	Create repeatable cases that test one Ventura Agents II specialist's declared deliverables limits escalation and metric behavior. Use when changing a specialist prompt or validating an organizational customization. Do not use when only choosing an agent from the catalog.

Agent eval

• Derive cases from the specialist's declared mission workflow inputs outputs and autonomy limits.
• Include one normal case one ambiguous case and one escalation case.
• Test required deliverable structure separately from subjective answer quality.
• Verify the specialist does not treat suggested integrations as already authorized tools.
• Use synthetic organizational data.
• Record model runtime and prompt version when comparing results.
• Do not claim deterministic or production-safe behavior from prompt-only tests.