# Reference Prompts The following annotated prompts illustrate the techniques and components described in this tutorial applied to common GLBNXT Workspace use cases. Each prompt is annotated to show which components and techniques are present and why they were chosen. *** #### Reference Prompt 1: Document Analysis and Briefing **Use case:** Analysing a lengthy internal report and producing an executive briefing. ``` You are a senior analyst supporting executive decision-making in a European enterprise organisation. Your output will be used directly in a board briefing. Write accordingly. Analyse the document provided and produce a structured briefing in the following format: ## Executive Summary Three sentences maximum. State the central finding and its strategic significance. ## Key Findings Up to five findings. Each as a single sentence followed by one sentence of supporting evidence. ## Risks and Considerations Up to three items. Each as a labelled risk with a one-sentence description. ## Recommended Next Actions Up to three actions. Each beginning with an imperative verb. Do not include background context about the document's origin or purpose. Do not speculate beyond what the document supports. If the document does not contain sufficient information to populate a section, write "Insufficient information" for that section rather than speculating. [paste document here] ``` **Annotations:** * Role prompting establishes the analytical frame and audience register * Output specification is precise and structural, reducing format variance * Constraints address the three most common failure modes for this task type (padding, speculation, hallucination) * Delimiter separates instructions from input data unambiguously * The "Insufficient information" instruction is a targeted anti-hallucination measure *** #### Reference Prompt 2: Few-Shot Classification **Use case:** Classifying support tickets by category and priority for routing. ``` Classify the following support ticket by category and priority using the system shown. Categories: Access, Billing, Technical, Feature Request, Other Priority: Critical (service blocked), High (significant impact), Medium (workaround available), Low (cosmetic or informational) Output format: Category: [category] | Priority: [priority] | Reason: [one sentence] Examples: Ticket: "I cannot log in to my account. The error says my credentials are invalid but I reset my password twice." Output: Category: Access | Priority: Critical | Reason: User is completely blocked from service access despite password reset attempts. Ticket: "Could you add dark mode to the interface? It would be easier on the eyes." Output: Category: Feature Request | Priority: Low | Reason: Cosmetic enhancement with no impact on functionality. Ticket: "My monthly invoice shows a charge for a plan I downgraded from last quarter." Output: Category: Billing | Priority: High | Reason: Incorrect billing has financial impact and requires manual investigation. Ticket to classify: [paste ticket here] ``` **Annotations:** * Few-shot examples demonstrate the classification logic for edge cases (ambiguous priority boundaries) * Output format is specified both verbally and demonstrated in examples, maximising format compliance * Three examples cover a representative spread of category and priority combinations * The examples are ordered from most to least urgent to prime the model's sense of the priority scale *** #### Reference Prompt 3: Chain-of-Thought Analysis **Use case:** Evaluating a proposed vendor contract for risk and compliance considerations. ``` You are a contracts analyst with expertise in European enterprise procurement and GDPR compliance. Review the contract extract below and reason through the following questions in order before producing your final assessment: Step 1: Identify all clauses that touch on data processing, data residency, or data transfer. Step 2: For each clause identified, assess whether it is compliant with GDPR requirements for a European enterprise. Note specific concerns. Step 3: Identify any clauses that create ambiguity about liability or enforcement jurisdiction. Step 4: Assess the overall risk level of the contract as: Low, Medium, or High. Step 5: Produce a list of up to five specific redline recommendations. Show your reasoning for each step before proceeding to the next. After completing all steps, produce a final summary section labelled "Assessment Summary" containing only the risk level and the redline recommendations. [paste contract extract here] ``` **Annotations:** * Chain-of-thought structure is explicit and ordered, preventing the model from compressing reasoning * Role prompt activates compliance-relevant analytical frameworks * The instruction to show reasoning before proceeding enforces the chain-of-thought discipline * The final summary section instruction separates the reasoning trace from the actionable output * Delimiter prevents the model from conflating contract language with instructions *** #### Reference Prompt 4: Iterative Document Drafting **Use case:** Drafting a structured policy document in multiple passes. **Pass 1: Structure generation** ``` You are a senior technical writer producing enterprise policy documentation for a European organisation operating under GDPR and ISO 27001 frameworks. Produce a document outline for a Data Retention Policy. The outline should include section headings and a one-sentence description of the content each section will contain. Do not write the full content yet. Produce the outline only. The policy must cover: scope, legal basis for retention, retention periods by data category, deletion procedures, exceptions and overrides, and review schedule. ``` **Pass 2: Section drafting (repeated per section)** ``` Using the outline produced in the previous step, draft the following section in full: [section name] Write at a level appropriate for a policy document: clear, precise, and unambiguous. Use plain English where possible without sacrificing precision. Do not cross-reference other sections that have not yet been drafted. Maintain consistency with the definitions and terminology established in prior sections. ``` **Pass 3: Review and consolidation** ``` Review the complete draft policy assembled from the sections produced above. Identify and correct: 1. Inconsistencies in terminology or defined terms across sections 2. Gaps where a commitment made in one section is not supported by another 3. Any language that is ambiguous about obligation (must vs. should vs. may) Produce a revised complete draft with corrections applied and a change log listing each modification made. ``` **Annotations:** * Instruction decomposition distributes a complex task across three distinct passes with clear handoffs * Each pass has a tightly scoped task instruction that prevents scope creep * Pass 2 includes a constraint specifically designed to prevent forward references to undrafted content * Pass 3 closes the loop with a structured review and a change log for auditability *** #### Reference Prompt 5: Adversarial Review **Use case:** Stress-testing a business proposal before stakeholder presentation. ``` You are a critical reviewer preparing a challenge brief for an investment committee. Your role is to argue against the proposal below, not to support it. Identify weaknesses, risks, and unsupported assumptions in the proposal. Structure your review as follows: ## Unsupported Assumptions List each assumption the proposal relies on that is not substantiated by evidence provided. For each, explain why it is an assumption rather than a demonstrated fact. ## Identified Risks List the three to five most significant risks. For each, describe the risk, its likelihood, and its potential impact if realised. ## Logical or Analytical Weaknesses Identify any reasoning errors, gaps in logic, or missing analysis. ## Questions the Committee Will Ask List the five most difficult questions a sceptical committee member would ask. Do not provide suggestions for improving the proposal. Your role in this prompt is to critique, not to advise. [paste proposal here] ``` **Annotations:** * Role prompting establishes an adversarial frame that directly counteracts sycophancy bias * The explicit directive not to improve the proposal enforces the adversarial scope * Structured output format ensures the critique is organised and complete rather than impressionistic * The "Questions the Committee Will Ask" section converts the critique into a practical preparation tool *** ### Closing Notes Prompt engineering is a skill with a measurable ceiling when approached casually and no practical ceiling when approached systematically. The techniques in this tutorial are not exhaustive. The field develops continuously, and the specific behaviour of any given model is subject to change with each new version. But the underlying principles, grounded in how transformer-based language models process text, are durable. The path from competent to expert is largely a matter of observation and iteration. Pay close attention to the responses you receive. Read failures as diagnostic signals. Build a library of tested prompts. Document what works and why. Over time, the patterns that separate high-quality prompts from mediocre ones will become intuitive, and the craft will become fast. In GLBNXT Workspace, every conversation you have and every prompt you write takes place within a sovereign European infrastructure designed to support exactly this kind of deep, iterative, context-rich work. The investment you make in prompt quality compounds within an environment where your context, your templates, and your configurations are yours to keep, refine, and share on your own terms. That is the foundation on which serious AI work is built. --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://docs.glbnxt.com/tutorials/guides/prompt-engineering-fundamentals/reference-prompts.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.