audit-foundation-models

Scan Foundation Models / Apple Intelligence integration code for issues that crash on unsupported devices, freeze the UI, or mishandle guardrail responses.

What This Command Does

Launches the foundation-models-auditor agent to flag the most common mistakes when integrating on-device language models — missing availability gates, a second request started while one is running, and brittle parsing of structured output.

What It Checks

1. Missing availability checks – LanguageModelSession use without isAvailable or device-capability gating
2. Concurrent requests – a button or action that can start a request while the session isResponding, which throws concurrentRequests
3. Manual JSON parsing – JSONDecoder against model output where @Generable would give compiler-checked structure
4. Missing error handling – no recovery for guardrailViolation, unsupportedLanguageOrLocale, assetsUnavailable
5. Session lifecycle – sessions created per-message instead of reused, or held across app states without checkpointing

Related Agent

• foundation-models-auditor – The agent that powers this command
• foundation-models-ref – Foundation Models reference