mastra-ai/mastra · error · MastraError

MAP_RESULTS_STEP_UNSUPPORTED_MODEL

MAP_RESULTS_STEP_UNSUPPORTED_MODEL

Error message

Tripwire handling requires a v2/v3 model

What it means

In the map-results step of agent streaming workflows, when the model raised a tripwire (guardrail violation), Mastra must re-read the model via capabilities.getModel to handle the tripwire consistently. If the resolved model is not a supported LanguageModelV2/V3 instance, tripwire handling cannot proceed, so this MastraError (USER category) is thrown.

Source

Thrown at packages/core/src/agent/workflows/prepare-stream/map-results-step.ts:127

          if (saveQueueManager && memoryData.thread?.id) {
            await saveQueueManager.flushMessages(messageList, memoryData.thread.id, memoryConfig);
          }
        }

        return options.onStepFinish?.({ ...props, runId });
      },
      ...(memoryData.tripwire && {
        tripwire: memoryData.tripwire,
      }),
    };

    // Check for tripwire and return early if triggered
    if (result.tripwire) {
      try {
        const agentModel = await capabilities.getModel({ requestContext: result.requestContext! });

        if (!isSupportedLanguageModel(agentModel)) {
          throw new MastraError({
            id: 'MAP_RESULTS_STEP_UNSUPPORTED_MODEL',
            domain: ErrorDomain.AGENT,
            category: ErrorCategory.USER,
            text: 'Tripwire handling requires a v2/v3 model',
          });
        }

        const modelOutput = await getModelOutputForTripwire<OUTPUT>({
          tripwire: memoryData.tripwire!,
          runId,
          ...createObservabilityContext({ currentSpan: agentSpan }),
          options: options,
          model: agentModel,
          messageList,
        });

        // End agent span with tripwire information after fallback completes
        agentSpan?.end({

View on GitHub (pinned to 75dd419e61)

Solutions

1. Return a genuine AI SDK LanguageModelV2/V3 model from your getModel capability implementation
2. Unwrap custom wrappers and pass the underlying provider model instance
3. Upgrade legacy v1 model adapters to v2/v3-compatible provider packages
4. Align @mastra/core, ai, and @ai-sdk/* provider versions so the model spec matches
5. Remove mocks/stubs that fail isSupportedLanguageModel in production paths

Example fix

// before
getModel: async () => myCustomWrapper,
// after
getModel: async () => myCustomWrapper.underlyingLanguageModelV3, // or the raw @ai-sdk provider model

import { isSupportedLanguageModel } from '@mastra/core/agent';
const model = await capabilities.getModel({ requestContext });
if (!isSupportedLanguageModel(model)) {
  throw new Error('Agent model must be an AI SDK LanguageModelV2/V3 for tripwire handling');
}

function isV2OrV3Model(m: unknown): boolean {
  return !!m && typeof m === 'object' && 'doStream' in m &&
    [2, 3].includes((m as { specificationVersion?: string }).specificationVersion as number) ||
    !!m && typeof m === 'object' && 'doStream' in m;
}

try {
  await workflow.start();
} catch (e) {
  if ((e as Error).message.includes('Tripwire handling requires a v2/v3 model')) {
    // replace the model adapter with a v2/v3-compatible one and rerun
  }
}

When it happens

Trigger: A stream finishes with result.tripwire === true while the agent's resolved model (possibly a custom wrapper, legacy v1 model, or non-standard provider object returned from getModel) fails isSupportedLanguageModel — i.e. not an AI SDK LanguageModelV2/V3.

Common situations: Using a custom model wrapper that doesn't implement the v2/v3 spec; passing a legacy model after upgrading Mastra/AI SDK; dynamic getModel returning a mocked or wrong-typed object in tests; mixing provider packages of mismatched major versions.

Related errors

• Processor tripwire triggered
• Aborted by processor
• Sub-agent ${agent.id} returned a v1 model but does not imple
• Sub-agent ${agent.id} returned a v1 model but does not imple
• Unsupported model version: ${(currentStep.model as { specifi

AI-assisted analysis of mastra-ai/mastra@75dd419e61 (2026-08-30). Data as JSON: /api/errors/715a248352b6447b. Report an issue: GitHub.