Skip to content

v0.7.0 - Machine-readable eval result contract

Choose a tag to compare

View all tags
@lindixu6-hash lindixu6-hash released this 16 Aug 21:23
v0.7.0
8a119e9

What shipped

  • Framework-neutral Eval Result JSON Schema linking observed runtime behavior to a known prompt-injection fixture.
  • Zero-dependency agentic-validate-results CLI for JSON objects, arrays, and JSONL.
  • Deterministic checks for required and unknown fields, duplicate result IDs, invalid outcomes, pass/violation consistency, timestamps, and broken fixture references.
  • Inert passing and failing examples with evidence, tool-trace, and policy-trace references.
  • English and Simplified Chinese adapter documentation with an explicit trust boundary: structural validation does not prove Agent safety.
  • 74 deterministic tests passing in CI.

中文说明

  • 新增框架无关的 Eval Result JSON Schema，将实际运行行为关联到已知提示注入 Fixture。
  • 新增零依赖 agentic-validate-results CLI，支持 JSON 对象、数组与 JSONL。
  • 校验必填/未知字段、重复结果 ID、非法结果、通过结果与违规记录的一致性、时间戳及断裂 Fixture 引用。
  • 提供无副作用的通过/失败示例，以及完整中文适配文档。
  • 明确信任边界：结构校验通过不等于模型或 Agent 安全。

Closes #11.

Assets 2
Loading