jsonschemaprofiles

A Go library and CLI tool for validating that JSON Schema documents conform to provider-specific structured-output restrictions.

This project validates schemas themselves—not JSON data instances—against profiles that capture the documented requirements of providers like OpenAI and Google Gemini.

Why?

LLM providers that support structured JSON output (OpenAI Structured Outputs, Gemini structured output) each accept only a subset of JSON Schema. A schema that works with one provider may be rejected by another. This library:

  • Validates schemas against provider profiles using a two-phase model (meta-schema + code checks)
  • Coerces schemas toward compliance with minimal, traceable changes
  • Provides a single library usable by multiple applications
  • Offers a CLI for quick validation and testing

Two-Phase Validation

  1. Phase 1 — Meta-schema validation: Validates the candidate schema against the profile’s embedded meta-schema file using JSON Schema Draft 2020-12.
  2. Phase 2 — Provider rules validation: Runs provider-specific checks not expressible in JSON Schema (limits, graph traversal, semantic invariants).

Available Profiles

Profile ID Description
OPENAI_202602 OpenAI Structured Outputs subset
GEMINI_202602 Gemini baseline structured output
GEMINI_202503 Gemini 2.0 with required propertyOrdering
MINIMAL_202602 Lowest common denominator across providers

Quick Start

As a Go Library 

import jsp "github.com/UnitVectorY-Labs/jsonschemaprofiles"

report, err := jsp.ValidateSchema(jsp.OPENAI_202602, schemaBytes, nil)
if err != nil {
    log.Fatal(err)
}
if !report.Valid {
    fmt.Println(report.Text())
}

As a CLI 

jsonschemaprofiles validate schema --profile OPENAI_202602 --in schema.json