The useful part

Find missing actors, unclear terms and untestable requirements.

Who it helps

People performing this specific task who can review and verify the result.

Put it to work

Prepare the template with your own context, inspect the preview, then copy it to your chosen model. The prompt uses 'Context grounding' by explicitly defining the input as untrusted data and requiring the model to cite specific passages. 'Role and audience framing' assigns the persona of a Senior QA Auditor to ensure a rigorous, professional tone suitable for product managers. 'Constraint specification' enforces strict rules: no inventing facts, returning only JSON with specific keys, and handling missing context by asking questions rather than hallucinating. 'Structured output' mandates a specific JSON schema for the ambiguity register. Finally, 'Self-critique and revision' is embedded in the instructions to verify that every identified ambiguity is traceable to the source text before finalizing the output.

Know the limits

This prompt relies entirely on the quality of the input text; if the brief is extremely vague or lacks any concrete details, the model may only generate questions without a substantive 'result'. It cannot resolve ambiguities that require external domain knowledge not present in the input or the model's training data without explicit clarification requests. The output is strictly JSON and cannot include conversational filler or markdown code blocks outside the JSON structure.

What we reviewed

Original template; fabricated example, output-contract and missing-input checks on qwen3.5-9b. Automated same-model critique, not independent human review.

Prompting techniques

The prompt uses 'Context grounding' by explicitly defining the input as untrusted data and requiring the model to cite specific passages. 'Role and audience framing' assigns the persona of a Senior QA Auditor to ensure a rigorous, professional tone suitable for product managers. 'Constraint specification' enforces strict rules: no inventing facts, returning only JSON with specific keys, and handling missing context by asking questions rather than hallucinating. 'Structured output' mandates a specific JSON schema for the ambiguity register. Finally, 'Self-critique and revision' is embedded in the instructions to verify that every identified ambiguity is traceable to the source text before finalizing the output.

Evidence and source statements

Evaluation scope

Fabricated example and missing-essential-context behavior · passed · qwen3.5-9b. This result applies to the declared task and model, not every possible use.