gh300-item-creator

name	gh300-item-creator
description	Generate GH-300 practice questions that feel like the real exam without copying it. Every item is grounded in current Microsoft Learn content, uses modern Copilot terminology, and follows Microsoft-style exam item rules (scenario-first, plausible distractors, no trick wording). Use when the user asks for practice questions, quiz items, or exam prep.

Skill: gh300.practice_questions.exam_realistic

Description: Generate GH-300 practice questions that feel like the real exam without copying it. Every item is grounded in current Microsoft Learn content, uses modern Copilot terminology, and follows Microsoft-style exam item rules.

Grounding

Required sources:

• Microsoft Learn (primary truth source for objectives and capabilities; access via the Microsoft Learn MCP server using microsoft_docs_search and microsoft_docs_fetch)
• Microsoft Learn samples (for syntax or command accuracy; access via microsoft_code_sample_search)

Study guide:

• Study guide for Exam GH-300 [Microsoft Learn]
• references/gh300-objectives.md for the full skills-measured list

Style

Microsoft style:

• Follow Microsoft sentence-style capitalization and UI-label rules.
• See references/style-guide.md for detailed writing rules.

Bundled assets

Use the skill-local bundle so this skill is reproducible and teachable as a package, not only a prompt.

• resources/source-pack.md
• resources/microsoft-voice-principles.md
• resources/item-quality-checklist.md
• scripts/validate-output.js

Guardrails

Exam integrity:

• Do not recreate or paraphrase real exam questions.
• Do not reference braindumps or leaked content.
• Write original scenarios and stems every time.

Terminology:

• Always use current GitHub Copilot and GitHub platform terminology.

Item quality:

• No contractions.
• Avoid negatives; if truly required, CAP + bold the negative word.
• Exactly 4 options (A-D) unless the requested item type explicitly differs.
• Exactly 1 correct answer unless the requested item type explicitly differs.
• No "all of the above" or "none of the above."
• Distractors must be plausible and real.

Answer choice randomization (non-negotiable)

You MUST randomize which letter (A, B, C, or D) is the correct answer for each question. Do not default to any single letter position. Across a set of questions, distribute the correct answer roughly evenly among A, B, C, and D.

Fictional company randomization (non-negotiable)

Use fictional company names from references/fictional-companies.md for scenario context. You MUST randomize the company selection -- do not default to Contoso for every scenario.

Workflow

1. Pull current GH-300 skill areas from references/gh300-objectives.md and choose a target objective.
2. Ground the intended correct behavior in Microsoft Learn using microsoft_docs_search first, then microsoft_docs_fetch if you need full page detail.
3. If the item touches command or settings specifics, invoke microsoft_code_sample_search where relevant.
4. Pick a random fictional company from references/fictional-companies.md and draft a workplace scenario stem.
5. Randomly assign the correct answer to A, B, C, or D. Write 1 correct answer and 3 plausible distractors.
6. Run a mutual exclusivity check on answer choices.
7. Run a terminology check.
8. Run a clarity check.
9. Run the checks in resources/item-quality-checklist.md and use scripts/validate-output.js logic as a final structure gate.
10. Prepare rationale internally but do not deliver it yet.

Recipe: responsible AI principle items

Use this recipe when the user asks for responsible AI, RAI, ethical AI, responsible usage, risks and limitations, or harms and mitigation questions. These map to the Use GitHub Copilot responsibly (15-20%) domain, specifically the Understand responsible AI principles and Validate and operate AI tools sub-groups.

Grounding (do this first, do not skip)

Ground every RAI item in the GitHub-Copilot-specific responsible AI content, not generic Azure AI or Copilot Studio responsible AI pages:

• Primary module: Responsible AI with GitHub Copilot (https://learn.microsoft.com/training/modules/responsible-ai-with-github-copilot/).
• The six principles unit: https://learn.microsoft.com/training/modules/responsible-ai-with-github-copilot/3-six-principles-of-responsible-ai.
• The mitigate AI risks unit: https://learn.microsoft.com/training/modules/responsible-ai-with-github-copilot/2-manage-ai-risks.

Use microsoft_docs_fetch on these URLs before writing, then cite the matching unit URL in Phase 2 references.

The six principles (the canonical set to test)

Microsoft and GitHub frame responsible AI around six principles. Test recognition AND application of each:

1. Fairness -- treat all groups equitably; watch for biased suggestions.
2. Reliability and safety -- perform consistently and safely; validate output before use.
3. Privacy and security -- protect data; respect content exclusions and data handling.
4. Inclusiveness -- work for people of all abilities and backgrounds.
5. Transparency -- make capabilities and limitations understandable.
6. Accountability -- humans remain responsible for AI-assisted outcomes.

RAI-specific distractor traps (use these to build plausible wrong answers)

• Confusing a principle with a similar-sounding one (for example, framing a bias scenario as transparency when it is fairness).
• Attributing a Copilot Studio or Azure AI Foundry control (content filters, RAI dashboard, Azure AI Content Safety) to GitHub Copilot. These are real features of OTHER products, which makes them strong distractors, but they are the wrong answer for GH-300.
• Treating Copilot output as authoritative and skipping human validation (violates reliability and safety plus accountability).
• Assuming Copilot removes the developer's responsibility for the code (violates accountability).

RAI item requirements

• The stem must present a workplace scenario where a responsible AI principle is at stake, then ask which principle applies OR which action upholds responsible use.
• At least one distractor must be a real responsible-AI control from a DIFFERENT Microsoft product (the wrong-exam trap above), so the item rewards knowing the GH-300 boundary.
• Rationale must name the specific principle and tie it back to a developer action with GitHub Copilot, not abstract ethics.

Delivery rules (non-negotiable)

When presenting a question to the user:

Phase 1 -- Question only:

• Show metadata, scenario stem, and choices (A-D).
• Do NOT include correct_answer, rationale, or references.
• End the message and wait for the user to reply.

Phase 2 -- Evaluation:

• After the user replies with their answer, show:
  • Whether they were correct or incorrect.
  • The correct answer letter.
  • Full rationale for every choice.
  • References (Microsoft Learn URLs).

If multiple questions were requested, repeat this Phase 1 / Phase 2 cycle for each question sequentially.

Output format

Phase 1 message (question only):

• metadata
  • exam: GH-300
  • skill_area: "<one of the GH-300 skill areas>"
  • objective: "<specific objective line>"
  • bloom: "<Remember|Understand|Apply|Analyze>"
  • difficulty: "<easy|medium|hard>"
• question
  • stem:
    • <Scenario + question. One decision.>
  • choices:
    • A: "<choice>"
    • B: "<choice>"
    • C: "<choice>"
    • D: "<choice>"

(Stop here. Wait for the user to answer.)

Phase 2 message (evaluation, after user replies):

• result: "<Correct! / Incorrect.> The correct answer is <A|B|C|D>."
• rationale:
  • A: "<2-sentence explanation>"
  • B: "<2-sentence explanation>"
  • C: "<2-sentence explanation>"
  • D: "<2-sentence explanation>"
• references:
  • "<Microsoft Learn URL 1>"
  • "<Microsoft Learn URL 2 if needed>"