math mentor persona

The problem

Standard LLMs tend to reveal final answers immediately, which prevents actual learning. Students often copy-paste solutions without understanding the underlying mathematical logic or identifying their specific conceptual gaps.

What it does

• Enforces a "hint ladder" system that requires student attempts before providing guidance.
• Diagnoses specific mathematical misconceptions by analyzing incorrect steps rather than just flagging them.
• Requires students to perform self-verification steps, like plugging values back in or sanity-checking magnitudes.
• Prompts students to explain the "why" behind key steps in their own words after solving a problem.

Why this beats prompting it yourself

Getting an LLM to stop giving away answers requires complex, persistent behavioral constraints that usually drift after a few exchanges. This skill implements a strict diagnostic framework and a persistence layer that maintains the "teacher" persona even during long tutoring sessions.

Use cases

• Guided homework support where the goal is mastery, not just completion.
• Diagnostic tutoring to find exactly where a student's algebra or calculus logic breaks down.
• Interactive practice sessions where the AI sets problems and provides step-by-step scaffolding.
• Self-study reinforcement for learners who want to be challenged rather than spoon-fed.

Known limitations

Will not provide final answers for graded work or exams. Requires the user to provide an initial attempt or a specific point of confusion to begin the interaction.