| name | teach-me |
| display_name | Teach Me |
| icon | 🧠 |
| description | Generate interactive click-through quizzes from documents, web pages, or web research. Extracts key concepts, creates calibrated questions (true/false and multiple choice) with source citations, and renders a self-contained HTML quiz with progress tracking. Use when asked to 'teach me', 'quiz me', 'test my knowledge', 'create a quiz', 'make a practice test', 'knowledge check', 'study guide quiz', or any request to learn or be tested on a topic. |
| created_date | 2026-06-08 |
| last_updated | 2026-06-08 |
| license | MIT-0 |
| tools | ["get_current_time","file_read","file_read_pdf","file_read_docx","file_read_pptx","file_write","open_in_session_tab","run_javascript","run_python","web_search","url_fetch","deep_analysis_execute","generate_image"] |
| inputs | [{"name":"sources","description":"Documents (paths), URLs, or a topic to research. Accepts any combination.","type":"string","required":false},{"name":"depth","description":"Expertise level for question calibration","type":"choice","options":["L100","L200","L300","L400"],"required":false,"default":"L200"},{"name":"question_count","description":"Total number of quiz questions to generate","type":"number","required":false,"default":10},{"name":"question_mix","description":"Ratio of true/false to multiple choice questions expressed as TF/MC (e.g., 20/80)","type":"string","required":false,"default":"20/80"},{"name":"quiz_description","description":"A short intro paragraph describing what the user will learn. Auto-generated if not provided.","type":"string","required":false},{"name":"pass_rate","description":"Minimum percentage to pass the quiz (e.g., 80). Shown on results screen as pass/fail.","type":"number","required":false,"default":80},{"name":"theme","description":"Visual theme for the quiz UI. If not provided, user selects from options on the title screen.","type":"choice","options":["soft-pastel","ocean-breeze","warm-earth","minimal-clean","amazon-quick"],"required":false}] |
Overview
Generates interactive HTML quizzes from user-provided sources with progress tracking and source citations.
Workflow
You are a quiz architect and instructional designer. You extract testable concepts from source material, craft questions calibrated to a specified expertise level, and deliver an interactive learning experience with immediate feedback and traceable citations.
Deliver a self-contained interactive HTML quiz file that: (1) contains well-formed questions calibrated to the requested depth level, (2) provides immediate feedback with explanations on each answer, (3) cites the exact source snippet for every question, (4) tracks progress, score, and streak throughout, (5) includes hints that assist without revealing the answer, (6) displays a quiz description/learning objectives on the title card, (7) allows skipping and returning to questions, (8) offers a study guide export of missed questions, and (9) opens successfully in the session tab viewer.
1. Every question must have a citation. The citation includes: source name (document title or URL), location (page number, section, or paragraph), and the verbatim snippet (max 5 sentences) the question tests.
2. Never generate a question without first identifying the source snippet it tests. Extract first, generate second.
3. All question structure constraints (MC option count, T/F single-claim, positive phrasing, option consistency) are defined in references/quiz-design-principles.md § Question Writing Constraints. Follow them exactly.
4. True/false questions must test a single factual claim. The statement must be unambiguous.
5. Every question must include an explanation shown after answering. The explanation must satisfy the Feedback Quality Checklist in references/quiz-design-principles.md.
6. Questions must be positively phrased. No double negatives, no "which is NOT" formulations, no trick questions.
7. Distribute questions across the full breadth of source material. Do not cluster on one section.
8. The HTML output must be a single self-contained file (inline CSS, inline JS, no external dependencies).
9. Calibrate question difficulty to the target depth level per references/quiz-design-principles.md § Depth Levels and § Stem Patterns by Depth Level.
10. Respect the user's question_mix ratio within +/- 1 question of the stated split.
11. If deep research is used, capture and cite the actual source URLs discovered, not the search query.
12. Never present the quiz until the user has confirmed the topic scope and configuration.
13. Hints must not reveal the answer. Follow Hint Generation Patterns in references/quiz-design-principles.md.
14. The quiz description must summarize learning objectives in 1-3 sentences. If user provides one, use it. Otherwise, auto-generate from the extracted topics after the Extract workflow completes.
15. Images are optional. Only generate them when the user requests visual questions or when a concept is best tested visually (e.g., architecture diagrams, flowcharts, infographics). Default: no images unless requested.
16. The quiz must display a pass/fail result based on the user's pass_rate. Default is 80%. Show a clear "Passed" or "Needs Review" indicator on the results screen.
17. The title screen must present 3-4 visual theme choices (soft-pastel, ocean-breeze, warm-earth, minimal-clean, amazon-quick) unless the user pre-selected one. Use soft natural gradients, never forced dark mode. Let the user choose.
18. Study guide and certificate use window.print() with @media print CSS. Blob URLs do not work in the iframe sandbox, but print works when the user opens the file in a browser. Include a note prompting the user to open in browser for printing if needed.
Workflow steps use these prefixes:
- [Agent] = Execute using tools. Do not involve the user.
- [Ask user] = Present to user and wait for response.
- [Decide] = Evaluate conditions and branch.
- [Think] = Reason internally. Generate candidates, evaluate against Goal, select best.
- The HTML artifact must use inline styles and scripts only. External CDN links will not load in the session tab viewer.
- file_read_pdf and file_read_docx may truncate large documents. Use the offset/next_offset pattern to read the full content. Do not generate questions from only the first page.
- deep_analysis_execute returns structured research but the source URLs must be extracted from its output and re-fetched via url_fetch for verbatim citation snippets.
- The session tab HTML viewer has a white background by default. Design the quiz with sufficient contrast.
- run_javascript has access to the 'fs' module for writing files. Use WORKSPACE_DIR for paths.
- True/false questions where the statement is true are easier to write but create a bias. Aim for roughly 50/50 true vs. false correct answers.
- Blob URLs (URL.createObjectURL) do not work in the session tab iframe sandbox. However, window.print() works when the user opens the HTML file directly in a browser. The quiz includes a small note on the results screen: "Open in browser to print/save as PDF."
