Errors

The SDK raises two error types, and which one you caught tells you where the problem is before you read the message.

import { EvalwrightApiError, EvalwrightClientError } from "@evalwright/sdk";

try {
  await client.prompts.resolve("support-reply", { vars });
} catch (err) {
  if (err instanceof EvalwrightApiError) {
    // The server answered, and the answer was a refusal.
    console.error(err.status, err.code, err.message, err.details);
  } else if (err instanceof EvalwrightClientError) {
    // The call never left your process: the client was constructed wrong.
  } else {
    // Network, DNS, timeout, abort — the platform's own errors, unwrapped.
  }
}

EvalwrightApiError

Property	Type	Description
status	number	HTTP status.
code	string	Stable machine-readable code — branch on this, not on the message.
message	string	Human-readable explanation.
details	unknown	Present when the server can say more, e.g. which variable was missing.

Status	Code	What it means
400	VALIDATION_ERROR	A required {{variable}} was not supplied.
401	UNAUTHORIZED	The key is missing, malformed, revoked — or not scoped to this prompt.
404	NOT_FOUND	No such prompt, or it has no production version and you did not pin one.

401 deliberately does not distinguish "revoked" from "out of scope": telling a caller which of the two it was tells them something about keys they do not hold.

EvalwrightClientError

Raised locally, before any request: a missing apiKey or a missing baseUrl. It means the client was built wrong, so retrying will not help — fix the construction.

Retrying

429 and 5xx arrive as EvalwrightApiError. Retry those with backoff. Do not retry 400, 401 or 404: the same request will be refused the same way, and a retry loop against 401 is how a key gets rate limited.