Primary sourcearXiv: Anthropic
View sourceEvent ·Official first-hand
Zep 推出 Agent Skills 服务:从验证轨迹自动编译智能体技能
Zep Agent Skills 是一项面向生产环境智能体的程序性记忆服务,把每次任务执行记录为带事件序列、结果和验证强度的 Trace。一个由 Reflector 和 Curator 组成的编译器只用通过验证的 Trace 生成版本化的 Skills,并通过扫描和准入门槛控制每个 Skill 版本的启用。该格式源自 Anthropic 发起并捐给 Agentic AI Foundation 的开放标准,宿主可通过 API 或 MCP 加载。在 τ³-bank banking、AppWorld 和 EnterpriseOps-Gym 上,教师模型 Trace 编译的 Skills 让三个较弱的执行模型在不同推理档位下通过率提升 4.5 到 18.8 点。在 ALFWorld 上,小模型运行编译的 Skills 收益达到甚至超过先前工作里持续微调的收益,且无需训练。
Current read
Zep 把智能体的成功操作记录编成可复用的技能包,弱模型直接拿来用通过率最多涨 18.8 点,还不用训练,做 Agent 的可以看看这套流程。
11 sources38° AI heatLast updated 2026/11/1 00:47:54
Evidence
11 sourcesRelated source 1shao__meng
View sourceRelated source 2Claude Code: GitHub Releases
View sourceRelated source 3宝玉
View sourceRelated source 4TestingCatalog
View sourceRelated source 5IT之家
View sourceRelated source 6Fierce Healthcare
View sourceRelated source 7HIT Consultant
View sourceRelated source 8claudedevs
View sourceRelated source 9lmarena.ai
View sourceRelated source 10Nathan Lambert
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.