Prompt Quality Audit Loop

By Juan Beltrán, Industrial B2B AI Transformation Executive.

Is this prompt good enough for business use? Use this before sharing a prompt with colleagues, agents, or a team workflow to check clarity, context, guardrails, output format, and failure modes. Prompt Quality Audit Loop Decision to make: Is this prompt good enough for business use? Decision owner: Prompt owner, knowledge manager, or governance lead. Working-session setup: - Timebox: 30 min working session - People in the room: Business owner, governance lead, system or data owner, and security or legal when required. - Preparation: Focused prep: gather one page of context and the decision you need. Context: [Paste your notes, excerpts, draft, meeting transcript, CRM fields, proposal text, public research, or examples here.] Context I should provide: - Prompt draft - Intended user - Tool environment - Allowed context - Desired output - Bad answer examples - Known risks Safety boundary: - Use only information I provide in this conversation. - Do not infer personal, confidential, regulated, pricing, customer, employee, or supplier facts. - If the material belongs in an approved enterprise environment, tell me before analyzing it. Instructions: Act as a prompt quality auditor for business use. Review the prompt below. Score clarity, context, output format, data safety, assumptions, and failure handling. Then rewrite it and provide realistic test cases. Run the session in this order: 0. Inspect the context. If a missing fact could materially change the recommendation, ask no more than five focused questions and wait. If I ask you to continue, mark each missing fact as unknown. 1. Identify the job: State what the prompt helps a person do and when it should not be used. 2. Audit context requirements: Check whether the prompt tells users what context to paste and what to avoid. 3. Audit output requirements: Check whether the answer format, quality bar, and review steps are explicit. 4. Add guardrails: Add safety boundaries, missing-context behavior, and assumptions language. 5. Create test cases: Write two or three realistic tests that reveal whether the prompt works. Evidence rules: - Separate supplied facts, interpretations, assumptions, and unknowns. - Reference the exact note, excerpt, metric, or example supporting every material claim. - Show the strongest credible counterargument to the recommendation. - Do not invent customer facts, benchmarks, financial numbers, policy approvals, or system access. - Do not turn missing evidence into a confident recommendation. - Keep the answer useful for Governance Lead. Output contract: A prompt quality score, failure risks, revised prompt, and test cases. Return: 1. BLUF: the decision, recommendation, or draft in plain language. 2. Evidence table: claim, supplied evidence, confidence, and gap. 3. Assumption ledger: what is assumed and how to verify it. 4. Counterargument: the strongest reason the recommendation may be wrong. 5. Decision record: decision status, accountable owner, next action, and due date or trigger. 6. Evidence still needed: only the gaps that could change the decision. 7. Stop condition: state when the work is complete and when it must pause. Evidence checklist: - User - Task - Context required - Output format - Guardrails - Failure mode - Test case Human operating ritual: - Test prompts with imperfect context. - Keep the bad-answer examples. - Do not publish prompts that hide judgment from the user. Do not use this loop when: Do not treat the output as legal, security, privacy, or policy approval. Stopping condition: Stop when the prompt can be tested by someone else and produces a reviewable output.

Canonical URL: https://juanbeltran.ch/operating-loops/prompt-quality-audit-loop