# Assessment Design

> Assessment Design with AI for L&D teams. Use this skill when an instructional designer or course developer wants to create quiz questions, knowledge checks, assessments, or test items using AI. Especially useful when going beyond simple MCQs — covering application-level questions, scenario-based items, distractors that actually work, and meaningful feedback. Trigger when someone asks for help wri…

- **Type:** Skill
- **Install:** `agentstack add skill-scibly-dev-skills-assessment-design`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [scibly-dev](https://agentstack.voostack.com/s/scibly-dev)
- **Installs:** 0
- **Category:** [Communication](https://agentstack.voostack.com/c/communication)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [scibly-dev](https://github.com/scibly-dev)
- **Source:** https://github.com/scibly-dev/skills/tree/main/skills/assessment-design

## Install

```sh
agentstack add skill-scibly-dev-skills-assessment-design
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Assessment Design with AI

Most AI-generated questions are too easy — they test recognition, not learning. Worse, a common AI failure mode is testing the wrong *thing* entirely: quizzing an exact percentage, date, or phrase pulled from the source text instead of whether the learner can actually use the underlying skill. A learner can memorize "73%" and still fail to apply the concept it came from; that's not transfer, it's trivia. This skill shows you how to use Bloom's taxonomy as a prompt lever to get application-level questions with quality distractors and feedback that actually teaches.

## Step 1 — Understand the assessment context

Ask the user:

1. **What's the topic?** Be specific — not "leadership", but "giving constructive feedback to a peer who missed a deadline".
2. **What Bloom's level?** If unsure:
   - *Remember / Understand*: learners recall or explain concepts → compliance or foundational knowledge
   - *Apply*: learners use knowledge in a new situation → skills and procedures
   - *Analyze / Evaluate*: learners break down situations or make judgments → complex decision-making
3. **Question format?** MCQ, scenario-based (situation + question), true/false with explanation, short answer, matching, drag-to-sequence.
4. **How many questions?** And for what purpose — formative check, end-of-course test, certification?
5. **What does the learner already know?** Helps calibrate how tricky the distractors should be.

## Step 2 — Build the assessment generation prompt

Construct this prompt:

```
ASSESSMENT GENERATION PROMPT

Role: You are an expert assessment designer and instructional design specialist.
Topic: [specific topic]
Target Bloom's level: [Remember / Understand / Apply / Analyze / Evaluate]
Learner: [role + experience level]
Question format: [MCQ / scenario-based / true-false / matching]
Number of questions: [N]

For each question:
- Write a stem that presents a realistic situation or judgment call — avoid "which of the following" where possible.
- [If MCQ] Write 4 answer options: 1 correct, 2 plausible distractors reflecting common mistakes, 1 tempting shortcut with a hidden flaw.
- [If scenario-based] Start with a 2-sentence situation before the question.
- Write feedback for each option: why it's correct or why it's a common mistake (2 sentences max per option).
- Tag each question with its Bloom's level.

Avoid: trivially obvious wrong answers, questions answerable by scanning the course text without thinking, trick questions, double negatives.
```

Run it if the user wants to see output.

## Step 3 — Quality check the output

Review generated questions against these criteria and flag issues:

| Check | What to look for |
|-------|-----------------|
| Distractors are plausible | Wrong answers reflect real misconceptions, not obvious nonsense |
| Stem is unambiguous | Learner knows exactly what's being asked without re-reading |
| Bloom's level is correct | Apply questions require doing something, not just recognizing a term |
| Tests transfer, not trivia | The question checks whether the learner can use the concept, not whether they memorized an exact figure, date, or phrase from the source material |
| Feedback teaches | Correct answer feedback explains WHY, not just "Correct!" |
| No giveaways | The right answer isn't the longest, most formal, or most cautious option |

Offer to revise questions that don't pass.

## Step 4 — Assessment prompt toolkit to take away

```
ASSESSMENT PROMPT TOOLKIT

--- For application-level MCQs ---
Write [N] multiple choice questions at the Apply level for [topic].
Learner: [role + experience level].
Each question: realistic situation stem, 4 options (1 correct, 2 plausible-wrong reflecting common mistakes, 1 tempting shortcut), feedback per option explaining why.
Avoid obvious distractors and trick questions.

--- For scenario-based questions ---
Write [N] scenario-based questions for [topic].
Each: 2-sentence realistic situation, then a decision or judgment question.
Options: [format]. Include per-option feedback.

--- For better distractors (fix existing questions) ---
I have this question: [paste question + correct answer].
Write 3 distractors that reflect genuine misconceptions a learner at [level] might have.
Each distractor: plausible, specific, reflects a real mistake — not obviously wrong.

--- For feedback that teaches ---
I have this question and answer: [paste].
Write feedback for each option that explains the reasoning, not just whether it's right or wrong.
Correct answer feedback: explain why this is the right call.
Wrong answer feedback: explain the common thinking error, then redirect.
```

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [scibly-dev](https://github.com/scibly-dev)
- **Source:** [scibly-dev/skills](https://github.com/scibly-dev/skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-scibly-dev-skills-assessment-design
- Seller: https://agentstack.voostack.com/s/scibly-dev
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.