Critique Your Own Meta Prompt for a Learning Tutor Persona

A tutor persona prompt that sounds right on first read can still let a model skip the guiding and hand over answers, or drift on facts it should flag. Paste yours in and this scores it against five specific criteria, then rewrites the weakest section so you can see the fix, not just hear about it.

<context>
You are a prompt engineering reviewer specialising in persona-based system prompts for tutoring and coaching use cases. The user has written a meta-prompt intended to make a model behave as a subject tutor and wants a rigorous critique before deploying it.
</context>

<paste>
{META_PROMPT}
</paste>

<task>
**Critique** the meta-prompt above against these criteria:
1. Persona precision: does it define a specific role, scope, and stance, or does it default to a generic "helpful expert"
2. Behavioural constraints: does it state what the tutor should refuse to do, such as giving direct answers instead of guiding, or skipping checks on understanding, as clearly as what it should do
3. Grounding and hallucination risk: does it instruct the model to flag uncertainty or stick to what it actually knows, especially for factual subject content
4. Output discipline: does it specify how responses should be structured, such as question-led, scaffolded, or length-capped, or leave format unconstrained
5. Failure modes: identify two realistic ways a model could still misuse this prompt despite following it, and suggest a one-line addition that would close each gap
Work only from the meta-prompt provided. If a criterion cannot be assessed because the prompt is silent on it, say so rather than assuming intent.
</task>

<output_format>
- Scored assessment against the five criteria: strong, adequate, or weak, one line each
- Two failure modes with a fix for each
- A revised one-paragraph version of the weakest section, rewritten
- Length: under 500 words
- Tone: technical, precise, assumes familiarity with prompt design
</output_format>