ETF Agent details isolated-context Subagent collaboration design
ETF Agent uses one Child execution mechanism for general research tasks, built-in Web, PDF and Python roles, six investment-research methods, account-specific profiles, a Research Reviewer and bounded Bull/Bear challenges. Its architecture isolates local work from the Main agent while preserving source permissions and final-answer ownership, and the report documents failures involving dropped source material, unidentified messages and interrupted execution. The article presents these as engineering behaviors, while stating that improved investment outcomes from adding specialists still require controlled validation.
The report offers a concrete reference for building multi-agent research systems that separate context, methods, permissions and accountability, while also showing why reliable handoffs and source provenance are harder than simply adding more experts.
ETF Agent 的 Subagent 已经不止 Web、PDF、Python 三种“工具专家”。 目前我们有通用 Child、内置专业角色、六种投资研究方法、账户自定义专家，以及按需参与的 Research Reviewer 和有界 Bull/Bear 质询。它们复用同一套执行机制，却不共享一团对话历史；专业方法可以不同，访问权限和最终回答责任并不会随角色名称转移。
这套体系最初只是为了减轻 Main 的上下文压力。一次 PDF 深读可能反复定位页码，一段 Python 研究可能多次修错。把局部过程隔离后，我们却碰到了更难的问题：专家真的看到了原始资料吗？成果真的进入了 Main 吗？多一个视角究竟改变了哪条研究结论？
本文想回答三个问题：我们现在具体支持哪些 Subagent，它们怎样在一项基金研究中工作；这些选择与业界框架有什么异同；哪些经验已经有工程证据，哪些收益仍值得继续验证。Web、PDF 和 Python 的具体实现留给专题文章，这里重点讲协作体系本身。
#ETF Agent#Anthropic#Microsoft Agent Framework#CrewAI#Deep Agents#LangChain