JSON Schema from samples: infer draft 2020-12 and validate data

What it does

Turn undocumented JSON (API responses, configs, exports) into a draft schema and check that data matches it.

• Merges every sample to determine types, nullability, required vs optional properties and array item types (for example, a field present in 23 of 45 objects is optional)
• Emits enums for strings with few distinct values, and formats (date-time, date, time, email, uuid, ipv4, ipv6, uri) only when every value matches
• Notes optional and mixed-type fields, and self-checks that all samples validate against the inferred schema
• Validates other data with a built-in validator and prints errors as file:line /json/pointer: reason
• Optional --bounds (observed min/max) and --strict (no additional properties)

How to use

python scripts/json_schema_tool.py infer samples/*.json --out schema.json --title Order
python scripts/json_schema_tool.py validate --schema schema.json new-batch.jsonl

Ask Claude, Codex, Hermes Agent or Cursor to "write a JSON Schema for this data"; it infers the schema, then reviews required fields and enums with you before finalizing.

Requirements

• Python 3.10+ (standard library only)

Caveats

• Inference only knows the samples it saw: with few samples, optional fields can look required and coincidental values can become enums.
• The built-in validator covers the common keywords only (if/then/else, unevaluatedProperties, remote $ref and others are not checked, and it says so). Use a full validator library when strict compliance matters.
• Data never leaves your machine. Enums may contain real values, so review them before publishing a schema.