v148 · AI · JavaScript

Safety Harness

14 edge-case tests for the Prompt API: API availability, model readiness, context overflow, prompt injection, empty input, very long input, structured output validation, and recovery paths. Run all or individually.

Checking Prompt API…

14Total

0Pass

0Fail

0Warn

Ready. Click "Run all" or run tests individually.

Test suite

// Production-safe Prompt API pattern
async function safePrompt(userInput, systemPrompt) {
  // 1. Feature-detect
  if (!window.ai?.languageModel) return { error: 'not-supported' };

  // 2. Check model availability
  const cap = await window.ai.languageModel.capabilities();
  if (cap.available === 'no') return { error: 'model-unavailable' };

  // 3. Guard empty / malicious input
  const sanitised = userInput.trim().slice(0, 8000);
  if (!sanitised) return { error: 'empty-input' };

  // 4. Limit context, stream, catch errors
  try {
    const session = await window.ai.languageModel.create({ systemPrompt });
    const stream = await session.promptStreaming(sanitised);
    let out = '';
    for await (const chunk of stream) out = chunk;
    return { result: out };
  } catch (e) {
    return { error: 'inference-error', message: e.message };
  }
}

implementation reference

Need the exact API surface, compatibility boundaries, errors, lifecycle, and source links? Read the matching gendn reference ↗

see also

implementation reference