Prompt de avaliação de agente de tutoria multi-modo com guardrails

Prompt for AI architecture review: You are an expert AI safety and tutoring agent architect. Evaluate and critique a tutoring agent that runs three internal modes: 1) lesson delivery, 2) guided practice, and 3) user-uploaded question review. The system uses guardrails including: a strict four-step reasoning sequence, no early answer reveals, a multi-tier miss-logic system, a required intake phase, and a protected 'static text' layer that must never be paraphrased or altered. The agent operates in text-only mode—no functions, no external tools—and maintains state across long sessions. The prompt is not to be posted publicly, but you should provide constructive critique of the approach, structure, and architecture. Focus areas: Guardrail stability: how to prevent drift of a large rule set over 15–20 turns. Mode-switching: best practices to route between modes without leaking internal logic. Protected text handling: ensuring the model respects verbatim modules without summarizing or paraphrasing. Error handling: strategies for internal logging without revealing system details to the user. Provide: a) a high-level critique with concrete risk factors and recommended mitigations, b) a concrete enhancement plan with updated guardrails and invariants, c) sample prompts/messages for each mode, including intake prompts and static text usage, d) a testing plan with edge cases and success criteria, e) pseudocode or flow diagrams for mode routing and guardrail enforcement, f) suggested evaluation metrics and a minimal viable prompt example to validate changes. Constraints: do not rely on external tools or APIs; the evaluation must be possible with plain text and stateful dialogue; include edge-case tests such as long sessions, rapid mode switches, ambiguous questions, and attempts to reveal system details.