跳转到内容

Tool ergonomics for agents

Tool ergonomics for agents 指工具是否以 agent 易于理解、选择、组合和高效调用的方式被设计出来。在当前 wiki 语境里,它强调的不是“工具能不能用”,而是“工具是否真正适合非确定性的 LLM agent 使用”。

  • 对 agent 来说,好的工具不是底层 API 的逐个包装,而是能顺着 agent 自然任务分解方式提供高价值动作的接口。
  • 这意味着工具设计要考虑的不只是功能覆盖,还包括:命名是否清晰、边界是否明确、返回信息是否高信号、上下文是否节省、错误提示是否可操作。
  • Anthropic 的经验表明,很多工程收益来自把若干低层操作折叠成一个更接近真实工作流的高层工具,例如 schedule_eventsearch_logsget_customer_context
  • The “think” tool- Enabling Claude to stop and think in complex tool use situations》补充了一个更少被讨论的方向:有些“好工具”并不负责执行外部动作,而是负责给 agent 一个结构化的思考节点。
  • namespacing 也是 ergonomics 的一部分。随着 agent 连接到越来越多 MCP server 和工具,清晰的前缀/分组能减少误用并降低描述进入上下文后的干扰。
  • 响应设计同样重要。像 concise / detailed 这样的 response format、分页、过滤、截断和高质量错误信息,都会直接影响 agent 的 token 使用效率和下游决策质量。
  • 这个概念与 Model Context Protocol (MCP) 相关,因为 MCP 让大规模工具接入成为常态;它也与 Agent scaffold 相关,因为工具定义、描述和返回格式本身就是 scaffold 的一部分。
  • Introducing advanced tool use on the Claude Developer Platform》补充了另一个关键维度:除了工具本身的 ergonomics,还要设计“工具如何被发现”“何时该程序化调用”“哪些调用约定需要 examples 才能学会”。
  • 这意味着 tool ergonomics 不只在单个工具定义里,也体现在整个 tool use workflow 的结构中。
  • 这也意味着 ergonomics 不只是在优化“行动工具”,有时也包括优化“认知工具”,例如显式的 Reasoning tools for agents

ℹ️ Conflict:

  • 面向 agent 的高层工具通常更高效,但也可能隐藏灵活性,限制 agent 进行非常规组合;因此“更 ergonomic”不总是等于“更通用”。
  • tool ergonomics 不能靠直觉判断。不同命名、返回格式或描述措辞的优劣,往往只能通过 evaluation 验证,而不是靠静态代码审查下结论。
  • Anthropic, “Writing effective tools for agents — with agents”, 2025-09-11: https://www.anthropic.com/engineering/writing-effective-tools-for-agents-with-agents
  • 原文摘录:[llm-wiki/raw/anthropic/Writing effective tools for agents — with agents](/raw/anthropic/Writing effective tools for agents — with agents.md)
  • Anthropic, “Introducing advanced tool use on the Claude Developer Platform”, 2025-11-24: https://www.anthropic.com/engineering/introducing-advanced-tool-use-on-the-claude-developer-platform
  • 原文摘录:[llm-wiki/raw/anthropic/Introducing advanced tool use on the Claude Developer Platform](/raw/anthropic/Introducing advanced tool use on the Claude Developer Platform.md)
  • Anthropic, “The “think” tool- Enabling Claude to stop and think in complex tool use situations”, 2025-03-20.
  • 原文摘录:[llm-wiki/raw/anthropic/The “think” tool- Enabling Claude to stop and think in complex tool use situations](/raw/anthropic/The “think” tool- Enabling Claude to stop and think in complex tool use situations.md)