Inline Kitty graphics now render LaTeX equations and Mermaid flowcharts directly in the artifact detail viewer, with Unicode ASCII fallback and left/right arrow panning. Interactive ask_question navigation lets users preview write-in answers before submitting, improving the questioning workflow. Dedicated rule token budgets allocate a 20,000-token limit per rule, preventing customizations from being evicted during long sessions. Model retry backoff is capped at 30 seconds, reducing long wait times after transient failures.