A specialist adds an action to a topic but the agent never uses it, even when a test case clearly calls for it. The action works when tested on its own. What should be examined first?
Select an answer to reveal the explanation.
Short Explanation
The agent picks tools by reading their labels. If the label is vague, the tool stays in the drawer. Nine times out of ten a never-invoked action has a description that could describe half the actions in the org.
Full Explanation
The reasoning engine selects an action by matching the current need against the action's description and the descriptions of its inputs, so a vague, generic or copied description is the leading cause of an action that is assigned but never chosen. Rewriting it to state plainly what the action does, when to use it and what each input means normally fixes the behaviour immediately. The company description is persona and organisational context and does not enumerate capabilities, so listing actions there would not influence selection. Creation order is not a selection criterion; the engine has no notion of first-added priority, which is the trap for anyone who imagines actions are evaluated as an ordered rule list. Trust Layer masking affects sensitive values passing to the model and can in edge cases affect what an input looks like, but it does not prevent an action from being considered, and it is the wrong place to start diagnosing a selection failure. Adjacent fact worth knowing: the same principle explains topic misrouting, since classification descriptions and action descriptions are both natural-language signals the engine reasons over, which is why both should be written for a machine reader rather than as internal documentation.