Function llmStage

Parameters

• config: {
      buildPrompt: (ctx: TStageContext) => { system: string; user: string };
      checkOutput?: (
          output: TOutput,
          ctx: TStageContext,
      ) => TLlmOutputCheckFailure | undefined;
      dependsOn: readonly TDepSpec[];
      id: string;
      maxOutputTokens?: number;
      model: string;
      outputSchema: TSchema;
      reasoningEffort?: TReasoningEffort;
      retry?: Partial<TRetryPolicy>;
      tools?: readonly TToolSpec[];
  }

  • buildPrompt: (ctx: TStageContext) => { system: string; user: string }

  • OptionalcheckOutput?: (output: TOutput, ctx: TStageContext) => TLlmOutputCheckFailure | undefined

    A check of the output's content, run once it matches the schema. A returned failure is handled exactly like a schema mismatch: the attempt is retried under the stage's retry policy (schema_validation must be in retryOn), and when the attempts run out the stage fails with that failure's code and message. It reads the stage context, so on the launch/complete path it runs only when completeStage is given the upstream outputs.

  • dependsOn: readonly TDepSpec[]

  • id: string

  • OptionalmaxOutputTokens?: number

  • model: string

  • outputSchema: TSchema

  • OptionalreasoningEffort?: TReasoningEffort

  • Optionalretry?: Partial<TRetryPolicy>

  • Optionaltools?: readonly TToolSpec[]