Foundation Models issues manifest as context window exceeded errors, guardrail violations, slow generation, availability failures, and unexpected output. Core principle 80% of Foundation Models problems stem from misunderstanding model capabilities (3B parameter device-scale model, not world knowledge), context limits (4096 tokens), or availability requirements—not framework bugs.
If you see ANY of these, suspect a Foundation Models misunderstanding, not framework breakage:
Critical distinction Foundation Models is a device-scale model (3B parameters) optimized for summarization, extraction, classification—NOT world knowledge or complex reasoning. Using it for the wrong task guarantees poor results.
Da utilizzare durante il debug dei problemi di Foundation Models: contesto superato, violazioni del guardrail, generazione lenta, problemi di disponibilità, linguaggio non supportato o output imprevisto. Diagnostica sistematica con difesa dalle crisi produttive. Fonte: megastep/codex-skills.