Files
Genarrative/rust/tests/agent-llm-eval.jsonl
kdletters 202279c6d9 新增独立 Agent Runtime Rust 工作区
新增 Core、Engine、Runtime、SQLite、Provider、MCP、Skill、Codex、CLI 与 DAG crate

补齐 OpenAI endpoint 配置、Provider 实例/协议路由和统一工具权限边界

加入持久化、lease、checkpoint、reconciliation、审批恢复与消息历史回归

加入独立 workspace CI、依赖边界、能力集和 Fake Agent 测试脚本

同步建设计划、TODO、架构、测试与验收文档
2026-09-06 17:44:54 +08:00

5 lines
1.9 KiB
JSON

{"schema_version":1,"id":"real-direct-token","provider":"openai","optional":true,"input":{"prompt":"只回复 AGENT_EVAL_OK,不要添加其它文字。","stream":true},"assert":{"status":"completed","final_text":{"mode":"contains_all","values":["AGENT_EVAL_OK"]},"steps":{"min":1,"max":3},"stream_events":{"min":1},"secrets_absent":true},"score":{"weights":{"status":0.3,"final_text":0.5,"streaming":0.2},"pass_threshold":0.9}}
{"schema_version":1,"id":"real-structured-answer","provider":"openai","optional":true,"input":{"prompt":"请只回复两个词:France 42。不要添加其它文字。","stream":true},"assert":{"status":"completed","final_text":{"mode":"contains_all","values":["France","42"]},"steps":{"min":1,"max":3},"stream_events":{"min":1},"secrets_absent":true},"score":{"weights":{"status":0.3,"final_text":0.5,"streaming":0.2},"pass_threshold":0.9}}
{"schema_version":1,"id":"real-stream-token","provider":"openai","optional":true,"input":{"prompt":"使用流式输出,只回复 STREAM_OK。","stream":true},"assert":{"status":"completed","final_text":{"mode":"contains_all","values":["STREAM_OK"]},"steps":{"min":1,"max":3},"stream_events":{"min":1},"secrets_absent":true},"score":{"weights":{"status":0.25,"final_text":0.55,"streaming":0.2},"pass_threshold":0.9}}
{"schema_version":1,"id":"real-echo-tool","provider":"openai","optional":true,"input":{"prompt":"必须调用 echo 工具一次,参数 text 必须是 AGENT_TOOL_INPUT。工具返回后只回复 AGENT_TOOL_DONE。不要猜测或跳过工具调用。","stream":true},"assert":{"status":"completed","final_text":{"mode":"contains_all","values":["AGENT_TOOL_DONE"]},"steps":{"min":2,"max":4},"stream_events":{"min":1},"tool_calls":{"required":[{"name":"echo","arguments":{"text":"AGENT_TOOL_INPUT"}}],"min_completed":1,"max_completed":1},"secrets_absent":true},"score":{"weights":{"status":0.25,"final_text":0.35,"streaming":0.1,"tool_calls":0.3},"pass_threshold":0.9}}