# Structured Outputs Between Nodes — OpenAI Agent Builder Workflows

Source: https://www.skillbyai.com/en/openai-agent-builder/d-schema

> Contracts between steps prevent silent breakage.

## Schemas as contracts

When node B depends on node A, the **schema** of A's output is a contract. Keep schemas small and specific: enumerations for categories, booleans for flags, numbers for amounts. Avoid free-text fields that later nodes must parse. When you change a schema, check every node that reads it, and version the workflow. Validate unusual values defensively: an enum blocks unknown categories, but a number may still be out of range.

## A schema-first design

Write the contract before the prompt.

```text
classifier output:
  category: enum[billing, technical, other]
  urgent:   boolean
  amount:   number | null      # parsed from text, null if absent
  summary:  string (<= 200 chars)

readers:
  route (if/else)        -> category, urgent, amount
  specialist agent       -> summary
  logging / analytics    -> all fields
```

## List the readers of each field

Before changing a field, look up which nodes read it; schema changes are the most common cause of broken workflows.

**Quiz:** Which field type is safest for a category used in routing?

- [ ] Free text
- [x] An enumeration of allowed values
- [ ] An image
- [ ] A random number

*Answer:* An enumeration of allowed values. Enums make invalid values impossible.