Works with the AI Agents you already use
Quiz Key Gate: Catch Answer-Key Errors in MCQ Quizzes
**AI tools write quizzes fast, and their answer keys break in the same ways every time.** The key says C but names option B's text.
$9
Quiz Key Gate: Catch Answer-Key Errors in MCQ Quizzes
Example session with this skill installed
Check my biology quiz and its answer key before I print it. (No file given, so the skill runs its bundled demo quiz.)
- Read your context and instructions
- Compiled the quiz key gate
Quiz Key Gate report
Verdict: BLOCK - the answer key has errors that would mis-mark students
File: biology-quiz-sample.md (text) - 12 questions, 12 with a usable key - CRITICAL 3 · HIGH 6 · MEDIUM 4 · LOW 0
1. Answer-key errors (CRITICAL)
| Q | Check | What is wrong | Fix |
|---|---|---|---|
| 4 | KEY-CONFLICT | the key disagrees with itself: answer line says A; answer-key section says B | Decide which answer is right and remove the other mark. |
| 7 | DUPLICATE-OPTION | options B and D are the same ('low concentration') and one of them is the key, so two options are right | Replace one of the two with a new plausible distractor. |
| 11 | DUPLICATE-QUESTION-KEY | same question as Q1 but keyed to a different answer ('golgi apparatus' vs 'mitochondrion') | Decide the right answer and delete one copy. |
2. Item-writing problems
| Q | Severity | Check | Evidence | Fix |
|---|---|---|---|---|
| 2 | HIGH | LONGEST-ANSWER | the key C is the longest option (88 characters vs at most 25 for the others): test-wise students pick the longest | Shorten the key or add the same level of detail to two distractors. |
| 3 | MEDIUM | NEGATIVE-STEM | the stem is negative ('not') and the word is not emphasised; students miss it under time pressure | Rephrase positively, or emphasise the word (NOT / bold / underline). |
| 5 | HIGH | GRAMMAR-CUE | the stem ends with 'an' and only the key fits it grammatically | End the stem without the article and put 'a'/'an' inside each option. |
| 8 | MEDIUM | ABSOLUTE-WORDS | 3 distractors use absolute words (always/never/only...) and the key does not | Remove the absolutes or use them evenly across options. |
| 9 | HIGH | OVERLAPPING-RANGES | ranges A (10-30%) and B (25-45%) overlap, so a value can sit in both | Make the ranges touch without overlapping (e.g. 10-19, 20-29). |
| 10 | HIGH | ALL-OF-THE-ABOVE-KEY | the key is 'All of the above': a student who spots two true options gets it without knowing the rest | Replace 'all of the above' with a single best answer, or turn the item into 'Select all that apply'. |
Connects securely to your tools. The creator never sees your data.
About this skill
AI tools write quizzes fast, and their answer keys break in the same ways every time. The key says C but names option B's text. Two options say the same thing and one is keyed. The same question appears twice with two different answers. The right answer is the only long one, or the only one that fits "an". Each costs a student a mark or hands the class a free point, and nobody notices until a student complains.
Quiz Key Gate reads the quiz and its key the way a careful exam reviewer would and returns a short report with a verdict (BLOCK / FIX FIRST / REVIEW / CLEAN), every problem by question number, the evidence, and a concrete fix.
What it checks
- Answer-key errors (CRITICAL): missing key, key letter the question does not have, letter + text that disagree ("C) Paris" when Paris is B), answer line vs key section conflicts, several keys on a single-answer item, "choose two" with the wrong count, duplicate keyed options, the same question keyed two ways, repeated option letters.
- Item-writing cues (HIGH/MEDIUM): longest-answer cue, a/an grammar cue, stem word echoed only in the key, absolutes only in distractors, qualified key, all-of-the-above as key, unemphasised NOT/EXCEPT/ليس/عدا, overlapping number ranges, near-duplicate options.
- Key balance: how keys spread over A-D, runs, repeating patterns, how often the right answer is also the longest.
- Answer-blind second solve (agent step): your agent solves the questions without looking at the key and lists disagreements as a clearly labelled second opinion for the teacher to decide.
- Optional rebalance for printed/text quizzes: a new copy with only the option order changed and the key updated; never touches wording, refuses while the key has errors.
Formats: Markdown/text as ChatGPT, Gemini or Word produce it (bold numbering, "Answer:" lines, key sections, tables, ✅ marks, options on one line), CSV incl. Google Forms-style sheets and the Kahoot and Quizizz templates, JSON (incl. Open Trivia style), Moodle GIFT. Arabic: abjad and alphabetical option letters, Arabic-Indic digits, diacritics and hamza forms normalised, right-to-left marks handled.
How it works: one standard-library Python script, deterministic (same file, same report), offline, never edits your file; outputs are new files only. Empty input runs a labelled demo on a bundled sample.
Limits: it checks structure and wording, not subject facts; the second solve is an AI opinion, not proof. Heuristic MEDIUM cues can fire on good items. Up to 10 options per question. Word/PDF quizzes must be saved as text first.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.
Checks structure and wording, not subject facts, difficulty or curriculum fit. Heuristic MEDIUM cues (word echo, absolutes, qualified key) can fire on well-written items. Up to 10 options per question; numbered questions and lettered options are expected in text files. Position-bias and longest-answer-rate checks need 8+ single-answer questions. Rebalance works on text quizzes only (use platform shuffling for Moodle/Kahoot/Quizizz/Forms) and does not carry over bold formatting or explanations written inside the answer-key section. Word/PDF must be saved as text first.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 days ago
- Software
Python 3.8+Terminal access
Permissions
File Scopes
No network access. Standard-library Python only.