安全评估 / NAVIN RESEARCH
Evaluate AI Agents End to End
Measure task success, policy compliance, tool behavior, efficiency, and recovery across complete trajectories.
Measure task success, policy compliance, tool behavior, efficiency, and recovery across complete trajectories.
简要回答: Agent evaluation must inspect more than the final answer. Score whether the task was completed, tools were chosen and called correctly, evidence supports claims, policies were respected, and the system stopped efficiently or recovered safely.
实施步骤
- Define outcome and process criteria from real tasks.
- Capture structured trajectories with redacted tool events.
- Use deterministic graders for actions and state changes.
- Add rubric or human review only for subjective dimensions.
验证清单
- Test repeated runs for variance.
- Include tool failures, stale data, and permission denial.
- Report results by task and risk slice.
常见错误
- Grading only the final prose.
- Rewarding longer trajectories.
- Letting the same model author every test and grade every result.
实际决策
先交付能够通过上述检查的最小设计。记录测试中使用的模型、提示词、工具、数据集与运行时版本，让后续变化可以被比较而不是被猜测。依赖供应商的限制、模型名称、价格或预览功能之前，请重新检查链接文档；这些细节变化往往快于本文的工程原则。
