Skip to main content

Configuration

Schemalint can be configured through CLI arguments, language-specific config files, or both.

CLI Arguments​

schemalint check [OPTIONS] [PATHS]...

OPTIONS:
  -p, --profile <PROFILE>    Provider profile ID or TOML path (can be specified multiple times)
  -f, --format <FORMAT>      Output format: human, json, sarif, gha, junit
  -o, --output <OUTPUT>      Write output to a file instead of stdout

For AI agents​

Agents should read the bundled, version-matched usage guide before running the CLI:

schemalint skills get core          # core loop, monorepo usage, exit codes, common fixes
schemalint skills get core --full   # plus the full command reference
schemalint skills list              # list bundled skills

schemalint --help points to the same guide at the top of its output.

Python (pyproject.toml)​

[tool.schemalint]
profiles = ["openai.so.2026-04-30"]
include = ["src/models/"]

Node.js (package.json)​

{
  "schemalint": {
    "profiles": ["openai.so.2026-04-30"],
    "include": ["src/models/"]
  }
}

How Zod schemas are selected (check-node)​

check-node lints every Zod object schema the files in --source can build without running a function. It does not matter whether the schema is declared or inline, exported or not, at module scope or inside a function body, or how it is spelled (z.object, z.strictObject, z.looseObject, a namespace import, zod/mini, a factory call). Barrel re-exports are followed to the definition, and a schema reached twice is reported once.

Two kinds of schema are not checked:

  • Schemas built from run-time input (they use a parameter, a local variable, or this) are skipped. No static tool can check them.
  • A schema Zod cannot convert to JSON Schema (Date, transforms, custom types) is reported as a warning and skipped. It does not fail the run.

Globs, files, and directories all follow the same rule. Use --source and --exclude to set the scope.

Every schema is checked against every selected profile. Without --profile, providers come from package.json dependencies: openai or @ai-sdk/openai selects OpenAI, @anthropic-ai/sdk or @ai-sdk/anthropic selects Anthropic, and both together select both. With neither, OpenAI is used.

Each schema is evaluated in isolation: from its own module, only the declarations and imports it depends on are loaded, so a module that also imports runtime-only bindings (cloudflare:workers, a database client, an HTTP app) evaluates cleanly. A local module the schema imports from is still loaded whole.