Simplified Schemas for Optional Fields
Tool schemas are now cleaner for optional fields by removing redundantanyOf: [{type}, {type: "null"}] constructs.
Background: How JSON Schema Handles Optional Fields
In JSON Schema, there are two distinct concepts:
Previously, our schemas used
anyOf with null type for all optional fields—even when the field wasn’t in required. This was redundant: if a field can be omitted, explicitly marking it as “accepts null” adds no value.
What Changed
For fields that are not in therequired array, the schema processing now:
- Removes the redundant
{type: "null"}fromanyOfarrays - Removes
default: nullsince it’s implied by being optional - Flattens single-type
anyOfto directtypedeclarations
Before vs After
For example, thepage_token field in GOOGLECALENDAR_LIST_CALENDARS (which is optional/not required):
Previous (verbose):
What’s Preserved
- Required nullable fields: Fields in the
requiredarray that accept null still useanyOfwith null type - Union types: Fields accepting multiple value types (e.g.,
string | number) retain their fullanyOfarray
Why This Matters
- Fewer tokens: Simpler schemas reduce token usage when tools are passed to LLMs
- Better compatibility: Some code generators and validators handle direct types better than
anyOfconstructs - Clearer semantics: Non-required fields don’t need explicit null type—being optional already implies they can be omitted