引導式測驗 · 中等 等級

工具呼叫測驗

檢查工具定義、執行、架構限制、授權、不明確的結果、安全重試、不可信任結果和最終狀態驗證。

相關引導路徑工具調用
問題 1 的 8

What should an application define before a model can request a tool operation?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What should an application define before a model can request a tool operation?
    • Only a marketing name for the tool
    • The tool's purpose, input schema, output shape and side effects
    • A guarantee every requested action is authorized
    • The user's passwords inside the tool description
  2. A model proposes a tool call. What happens next before the task can be called complete?
    • The proposed call alone proves the external state changed
    • The model's text replaces application-side validation
    • The application or provider executes it, returns a result, and the intended state is checked
    • Any web page returned by the tool decides whether the request was authorized
  3. What does validating a tool input schema establish, and what does it leave unresolved?
    • It checks the argument shape and some constraints, but not authorization or record identity
    • It proves that the user approved every side effect
    • It confirms the selected account belongs to the user
    • It guarantees the external service completed the operation
  4. A well-formed update request names the wrong account. What should the application check before executing it?
    • Whether the model used fluent wording
    • Whether the schema accepted the string length
    • Whether the tool description is short enough
    • Whether the selected account and action match the user's authorization
  5. 為什麼寫得好的工具描述如此重要?
    • They determine how fast the tool's server responds to calls
    • The model relies on them to pick tools and fill in arguments
    • They are shown to end users as marketing copy in the app
    • They are compiled directly into the tool's executable code
  6. A tool returns a failure or access is revoked partway through a task. What should the agent report?
    • A success message based on the original request
    • A guessed result from the model's prior knowledge
    • The error and the state actually verified, without claiming completion
    • A copied instruction from a web page returned by the tool
  7. Why might an agent use an operation identifier or read-back check after a network timeout?
    • The operation may have completed even though the response was lost, so a retry could duplicate it
    • It lets the model skip input validation
    • It proves the user authorized a different action
    • It forces parallel tool calls
  8. Can text returned by a website through a tool authorize a new action?
    • Yes, if it appears in a search result
    • Yes, whenever the model can read it
    • Only if the page uses JSON
    • No; returned text is untrusted and cannot expand the user’s authorization