Profiles

A profile defines what a specific LLM provider supports, restricts, or forbids in JSON Schema. Schemalint ships with built-in profiles for:

• OpenAI Structured Outputs
• Anthropic Structured Outputs

How Profiles Work

Each profile is a TOML file that declares:

• Keyword behavior: whether a JSON Schema keyword is allowed, forbidden, warned, stripped, or unknown
• Value restrictions: allowed values for keywords like format and additionalProperties
• Structural limits: constraints on schema structure (max depth, max properties, etc.)

The linter compares your schema against the profile and emits diagnostics wherever a keyword or pattern would be rejected or altered by the provider's API.

Creating a Custom Profile

You can create a custom profile file and pass it with --profile path/to/custom.toml:

name = "my-provider.v1"
version = "1.0"
code_prefix = "MYP"

type = "allow"
properties = "allow"
additionalProperties = { kind = "restricted", allowed = [false] }

[structural]
require_object_root = true
require_additional_properties_false = false

Custom profiles inherit all Class A and Class B rules from the rule engine — the profile data alone determines which rules fire and at what severity.