Purpose-driven
Define the legitimate goal, audience, context, output, criteria, evidence, and verification requirements.
Assignment overview
The Prompt Engineering Portfolio assesses your ability to move beyond casual prompt wording and design complete, purpose-driven AI-supported tasks. You will diagnose an under-specified request, redesign prompts with PCAFE-V, compare outputs using observable criteria, minimize context and protected information, design a staged SCOPE workflow, and allocate responsibility through PARTNER.
The portfolio should show both process and judgment. Include initial and revised prompts, selected outputs or supplied records, evaluation evidence, corrections, source and information boundaries, human review gates, and a concise rationale for your final design.
Comparative
Show how and why a redesign improved, failed to improve, or changed the result using observable criteria.
Human-governed
Assign suitable AI and human roles, preserve meaningful review, and define escalation and stop conditions.
Portfolio principle: A stronger prompt does not guarantee an accurate or responsible result. The portfolio must show evaluation, verification, information boundaries, and accountable human decisions.
What the portfolio should demonstrate
- Diagnose missing purpose, context, audience, format, criteria, evidence, and verification requirements.
- Apply PCAFE-V to produce a clear, bounded, and evaluable prompt.
- Identify assumptions, ambiguity, uncertainty, and inappropriate requests for inference.
- Minimize context while preserving the legitimate purpose.
- Protect personal, confidential, proprietary, privileged, sensitive, and culturally governed information.
- Compare outputs using criteria defined before selecting a preferred result.
- Separate fluency, usefulness, and formatting from reliability and evidence.
- Decompose a complex task into SCOPE stages, sources, prompts, outputs, and decision gates.
- Allocate work among the user, AI system, specialist reviewer, and final decision-maker using PARTNER.
- Recognize automation bias and responsibility laundering.
- Document prompts, outputs, verification, corrections, and human decisions.
- Explain limitations, residual risk, escalation, and stop conditions.
Required portfolio components
The official Assignment 2 instructions determine the final number of examples, length, and file format. The components below reflect the approved Module 2 activities and should appear in the portfolio.
Weak-prompt diagnosis
Identify what is missing or unsuitable in an under-specified request. Address purpose, context, audience, format, criteria, source expectations, verification, assumptions, ambiguity, and information risk.
Evidence: annotated original prompt and diagnosis table.
PCAFE-V prompt redesign
Transform the original request into a complete, purpose-driven prompt with clear boundaries and evaluation requirements.
Evidence: revised prompt and design rationale.
Prompt and output comparison
Compare initial and revised results using observable criteria. Explain improvements, trade-offs, remaining errors, and why fluency alone is insufficient.
Evidence: selected outputs, comparison matrix, and conclusion.
Context-minimization exercise
Identify necessary, unnecessary, sensitive, and inappropriate information. Redesign the context to use the minimum information required.
Evidence: before-and-after context and information rationale.
SCOPE sequence design
Decompose a complex task into stages with source boundaries, intermediate outputs, evaluation gates, revision decisions, and stop conditions.
Evidence: staged workflow and text alternative.
PARTNER responsibility matrix
Allocate tasks among the user, AI system, qualified reviewer, and final decision-maker. Address automation bias, escalation, human approval, and accountability.
Evidence: responsibility matrix and justification.
Use one coherent use case where possible: Reusing a bounded, low-risk case across components makes it easier to show development and integration.
Recommended production process
- Define
Select a low-risk task, legitimate purpose, audience, output, and success criteria. - Design
Diagnose the request, apply PCAFE-V, minimize context, and identify sources. - Test
Generate or inspect supplied outputs and compare them using pre-defined criteria. - Govern
Build the SCOPE workflow, PARTNER matrix, verification record, and final rationale.
Show the recommended evidence chain
- Original task request and weak prompt
- Prompt diagnosis
- PCAFE-V redesign
- Initial and revised outputs or supplied equivalents
- Pre-defined evaluation criteria
- Output comparison and identified errors
- Context-minimization record
- SCOPE sequence and review gates
- PARTNER responsibility matrix
- Verification, correction, residual-risk, and stop record
- Final reflective rationale
- AI-use disclosure
Evidence and evaluation requirements
Evaluate outputs for
- Alignment with purpose and audience
- Completeness and relevance
- Accuracy and supported claims
- Appropriate assumptions and uncertainty
- Format and usability
- Fairness and accessibility
- Privacy and information protection
- Source quality and traceability
- Need for revision or qualified review
Document your process
- System or supplied record used
- Date and version, if available
- Exact prompt or instruction
- Information and sources supplied
- Output retained
- Evaluation criteria
- Verification performed
- Corrections made
- Human decisions and approval
- Material included in the portfolio
Verification rule: Do not treat generated citations, quotations, names, statistics, calculations, policies, or technical claims as verified. Check consequential content against suitable authoritative sources.
Participation pathways
Option A: Approved AI tool
Use an approved system with synthetic, public, supplied, or authorized information for a bounded, low-risk task.
Option B: Supplied records
Use supplied prompts, outputs, context records, source sets, workflow stages, and responsibility scenarios.
Option C: Non-AI analysis
Complete the same diagnosis, redesign, comparison, workflow, and responsibility analysis without live AI use.
All pathways should demonstrate the same learning outcomes and be evaluated using the same standards.
Responsible use, authorship, and boundaries
Required boundaries
- Do not upload real student, employee, client, patient, customer, community, or confidential records.
- Do not use the assignment for health, legal, financial, safety, employment, disciplinary, or other consequential advice.
- Do not connect email, files, accounts, payments, devices, calendars, or external-action tools.
- Do not test prompt injection or security weaknesses against live systems.
- Do not infer identity, protected characteristics, credibility, diagnosis, intent, or risk about real people.
Student responsibility
- You must author the diagnosis, evaluation, rationale, workflow, and conclusions.
- You must understand and be able to explain every included prompt and decision.
- Document material AI assistance accurately.
- Verify factual and source-based claims.
- Provide text alternatives for visual workflows.
- Stop and seek guidance when authority or evidence is inadequate.
Suggested portfolio organization
- Title page and portfolio overview
- Use-case purpose, boundaries, audience, and information classification
- Weak-prompt diagnosis
- PCAFE-V redesign and rationale
- Output comparison and evaluation evidence
- Context-minimization analysis
- SCOPE sequence and review gates
- PARTNER responsibility matrix
- Verification, corrections, limitations, residual risk, and stop conditions
- Final synthesis explaining what the portfolio demonstrates
- AI-use disclosure
- References and permitted appendices
Use the official instructions if they prescribe a different structure.
Assignment 2 resources and submission
Assessment details
Assessment: Assignment 2, Prompt Engineering Portfolio
Weight: 15%
Recommended submission: End of Module 2
Official Moodle due date and time: ____________________
Required length and file format: ____________________
Before submitting
- Review the official instructions and rubric.
- Confirm all required components are included.
- Verify claims, citations, and references.
- Remove sensitive or unauthorized information.
- Complete the AI-use disclosure.
- Check accessibility and file naming.
- Upload the correct file and verify submission status.
Course developer: replace the placeholder anchors with the final Moodle URLs or companion HTML filenames.