Real tools have three or four parameters

The first tool was easy. one input, one output. Reality is uglier: most useful functions take 2-5 parameters, some with limited values (enum), others with specific formats (dates, IDs).

Your job here is to define assign_shift: an operation that mutates the shift roster. Three parameters:

• crewmate_alias. string. The crewmate's alias.
• shift_id. string. Specific format: shift-YYYY-MM-DD-NN.
• role. string. Only five valid values: lead, engineer, medic, comms, support.

What's new: enum

When a parameter accepts only a finite set of values, use enum:

"role": {
  "type": "string",
  "enum": ["lead", "engineer", "medic", "comms", "support"],
  "description": "Role during the shift."
}

With enum, the model can't invent values that don't exist ("captain", "doctor", "comms_lead"). The system rejects out-of-enum values and the model quickly learns the valid set.

What's new: parameter descriptions

Each parameter has its own description. For generic types (string, number), the description is what tells the model what to put in there. Example: shift_id is a string, but the model won't know the format unless you say so.

Rule of thumb: if you could pass the parameter wrong (weird format, invalid value), add its description.

Mutating vs read-only operations

lookup_crewmate is read-only. assign_shift mutates state. The tool description must make it clear which one is which. the model decides differently for operations that change things.