ml-course-interactive-learning-assistant

name	ml-course-interactive-learning-assistant
description	Use when creating an interactive review for an ML-course lecture and its course-specific source, safety, and publishing rules are needed.

ML Course Interactive Learning Assistant

This is the ML-course adapter for interactive-learning-experience-builder. Use that portable core for the experience specification, content contract, single-file rendering, accessibility requirements, and reusable validation. This adapter supplies only stable ML-course constraints.

Course context

• Read repository instructions and relevant contributor guidance before selecting sources.
• Resolve the requested lecture slug through lectures/index.yaml.
• Store the grounded payload at lecture_experiences/content/<lecture_slug>.json and the generated review at lecture_experiences/<lecture_slug>/index.html.
• The committed offline HTML is the source for the Pages copy. Build tooling copies it to site/_build/demos/<lecture_slug>/index.html; do not maintain a second demo under site/.

Source and safety policy

Ground claims in this order:

1. lectures/<lecture_slug>/lecture_notes.md;
2. lectures/<lecture_slug>/README.md and links.yaml;
3. public lecture_examples/ material;
4. the public practical README and student notebook;
5. relevant okf/ concepts as read-only supporting sources.

Do not modify okf/. Do not use private solutions, teacher notebooks, answer keys, grading data, unpublished drafts, or untracked quiz workbooks. Name every public course source used in the payload. The course generation wrapper enforces those restricted path categories and rejects lectures/ sources outside the selected lecture while permitting okf/ as read-only supporting context. Its source allowlist accepts only canonical lowercase lectures/<lecture_slug>/... and okf/... repository paths. Other relative roots, URLs, knowledge-base identifiers, and case-variant roots are rejected even though the portable core can support them in unrelated repositories.

Recurring generation workflow

1. Ask the learner for four defaults: Foundations, Applied, or Challenge quiz depth; focus-friendly mode on or off; color-blind-safe palette on or off; and funny topic-related break prompts on or off.

2. Create the short experience specification required by interactive-learning-experience-builder, then ground the payload using the course source order above. For every visualization, name the lecture objective it teaches and record it as a named lecture objective in the experience specification. Do not repeat a generic chart set across lectures; reject a repeated generic chart set when a supported semantic visualization type better represents the objective. Before publishing, exercise every visualization control and both palette modes.

3. Include the three foundations, applied, and challenge quiz banks with exactly 10 questions each. Embed break prompts even when their initial display setting is off.

4. Generate through the thin course-policy wrapper, which delegates rendering to the portable core only after source validation:

   uv run python .agents/skills/ml-course-interactive-learning-assistant/scripts/generate_course_learning_experience.py `
     --lecture-slug <lecture_slug> `
     --content lecture_experiences/content/<lecture_slug>.json `
     --template .agents/skills/interactive-learning-experience-builder/assets/learning-experience-template.html `
     --output lecture_experiences/<lecture_slug>/index.html
   

5. Validate with the portable core:

   uv run python .agents/skills/interactive-learning-experience-builder/scripts/validate_learning_experience.py `
     lecture_experiences/<lecture_slug>/index.html
   

6. Open the generated index.html directly through file://. Verify all four learner settings, visualizations and their fallbacks, answer review and whole-quiz Retry, keyboard navigation, visible focus, reduced motion, and storage fallback when storage is disabled. The core remains responsible for deterministic single-file output, static no-JavaScript explanations and quiz review, and accessible chart fallbacks.

Course checks and deployment

Before committing a lecture review, run the focused experience tests and the course preview build:

uv run pytest tests/test_eda_lecture_experience.py tests/test_interactive_learning_assistant_skill.py tests/test_interactive_learning_assistant_docs.py tests/test_interactive_learning_experience_builder_skill.py tests/test_learning_experience_portability.py -q
uv run python tools/build_textbook_preview.py

After merging to main, verify GitHub Actions and the deployed Pages review in the public student repository.