[QTI] Text entry question with an image in its prompt fails to save #6267

❌ This issue is not open for contribution. Visit Contributing guidelines to learn about the contributing process and how to find suitable issues.

Target branch: unstable

Observed behavior

Adding an image or small text to the prompt of a text entry question (numeric, text entry, free response) produces an item the server rejects, so the question doesn't save.

• TipTap emits images and small text as top-level <img> and <small>.
• buildTextEntryInteractionXML writes the prompt directly into <qti-item-body>, which the QTI schema only allows to hold block elements.
• The sync API runs validate_qti_item on every QTI write and rejects the item.

Inline choice has the same constraint and already handles it with BLOCK_TAGS / wrapInlineRuns in interactions/inlineChoice/parse.js, but only for itself. Every other interaction writes TipTap content into qti-prompt, qti-simple-choice, qti-simple-associable-choice or qti-html-content, all of which accept it.

Errors and logs

Schema check on an item body with a top-level image, as the text entry builder writes it:

Element '{http://www.imsglobal.org/xsd/imsqtiasi_v3p0}img': This element is not expected.

Expected behavior

Any TipTap content in a text entry prompt saves and reopens unchanged. Every interaction that writes authored HTML directly into <qti-item-body> goes through a single shared helper, instead of each one handling the rule itself:

• The inline choice wrapping moves out of inlineChoice/parse.js into a shared helper in serialization/xml.js that turns an HTML fragment into a list of block elements for <qti-item-body>.
• Builders call it where they place content, before setting any markers: inline choice for its prompt and passage (its data-studio-prompt marker must stay on a top-level element), text entry for its prompt.

User-facing consequences

Authors can't add an image or small text to the prompt of a numeric, text entry or free response question. The save fails with a sync error.

Steps to reproduce

1. Open an exercise and add a numeric or text entry question.
2. Add an image to the question prompt.
3. Save. The item is rejected by the sync API's QTI validation.

Acceptance Criteria

• A text entry prompt containing an image, small text, or both saves, passes validate_qti_item, and reopens with the same content
• Inline choice keeps its current behavior, including prompt/passage separation for prompts with images or small text
• Inline choice and text entry use the same helper from serialization/xml.js; inlineChoice/parse.js no longer defines its own
• Tests cover a text entry prompt with an image and with small text, built and parsed back

Context

Alternatives considered:

• Wrap in assembleItemXml: covers every interaction without builders calling anything, but it runs after builders set their markers. Inline choice would tag a bare <img data-studio-prompt>, assembly would wrap it in an unmarked <div>, and on reopen the image would move from the prompt to the passage.
• Make TipTap emit only block elements at the root: only fixes TipTap's output, not stored or imported XML that reaches the inline choice parser. It also changes the stored markup of every rich-text field where a top-level <img> is already valid, and Image and Small would have to parse both the old and new forms permanently.

AI usage

I used Claude Code to check TipTap's output and every interaction's content slot against the server QTI schema, compare the alternatives, and draft this issue. I reviewed the findings and edited the draft. The save failure is based on the schema check, not reproduced in the app.