Posts tagged with "Evaluation"

Wrapping device operations as tools an agent can call: the three layers, why a tool description is an inference rule rather than documentation, and the two shells a deeplink has to survive.

把裝置操作包成 agent 能呼叫的工具，中間踩到的兩層 shell、被悄悄蓋掉的 timeout，以及為什麼 tool description 寫完還不夠。