feat(quizzes): add AI-graded short answers and improve question quality

[mahruschow]

Quizzes currently support only multiple-choice responses, and generated distractors can make the correct answer obvious. Adds a Short answer creation option, model answers and grading criteria, and server-side AI grading with saved feedback. Strengthens multiple-choice authoring guidance to request realistic misconceptions, reasoning questions, and comparable option wording.

Short-answer submissions enforce workspace access and existing AI usage limits. Grading runs outside database transactions, validates structured output, and saves a first-submission result. Edits invalidate old grades; concurrent edits or progress resets reject stale results. Failed grading preserves the response in the tab for retry. Mixed quizzes are supported by progress, missed-question sessions, AI reads, search, and Markdown exports.

Validation:

• 53 focused tests passed, including short-answer content, edits, grading failures, usage limits, persistence races, and AI read contracts.
• TypeScript, scoped formatting/lint checks, and git diff --check passed.
• Production build passed.

Draft: local interactive verification is blocked by PostgreSQL connection refusal on localhost:5432. Live model grading quality and desktop/mobile UI interaction still need verification with the team's development environment and AI Gateway credentials. Persistence race tests mock the database; real PostgreSQL concurrency should also be exercised before marking ready.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

---

_{Comment @coderabbitai help to get the list of available commands.}

[cubic-dev-ai]

All reported issues were addressed across 3 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[greptile-apps]

Confidence Score: 4/5

[Medium risk] Improves quiz generation prompts and validation descriptions.

Neither issue blocks merging. The dialog prompt should be adjusted so topic-only and true/false requests receive consistent guidance.

Findings

1.  Allow topic-only quizzes ▶
2.  Align true/false guidance ▶

Summary

The PR strengthens quiz-writing guidance in the quiz tool and Create Quiz dialog, and documents quiz behavior. The dialog prompt needs two non-blocking adjustments: it asks for workspace source material even when a request supplies only a topic, and it gives true/false requests distractor guidance that conflicts with the quiz tool.

_{Reviews (1) · Last reviewed commit: "fix(quizzes): strengthen question and di..."}

[greptile-apps]

Allow topic-only quizzes

If a workspace has no relevant source material, the dialog still asks the AI to use it. That instruction cannot be followed for a topic-only request and may lead to a clarification instead of the requested quiz. Qualify the instruction for cases where relevant material is available. This is a non-blocking prompt concern.

Knowledge Base Used: Workspace learning tools

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

Artifacts

Executable Chromium comparison command

• This script rendered the parent or PR dialog and submitted a topic-only request, capturing the resulting composer queue.

Rendered dialog and composer queue fixture

• This fixture mounted the real dialog and displayed its queued prompt without supplying workspace sources.

Empty-workspace browser fixture

• This page hosted the dialog with no source files, quotes, or attachments.

Parent revision submission log

• The recorded command exited 0 and captured a queued topic-only prompt without the source-material instruction.

▶ Parent revision dialog and topic-only submission

• Chromium submitted the topic-only request through the parent dialog, showing the original prompt in the queue.

Parent revision queued prompt

• The captured browser frame shows the submitted topic-only prompt without a source-material instruction.

PR revision submission log

• The recorded command exited 0 and captured the same queued request with the added source-material instruction.

▶ PR revision dialog and topic-only submission

• Chromium submitted the same request through the PR dialog, showing the changed prompt in the queue.

PR revision queued prompt

• The captured browser frame shows the added source-material instruction despite the fixture having no sources.

Chromium validation command source

• The executed script renders both dialog versions, submits the same topic, and checks the queued prompts against the tool schema; it documents the comparison.

Rendered dialog validation source

• The executed page mounts the real dialog and displays the composer queue and live schema description; it shows what the browser test exercised.

Pre-PR dialog source used for comparison

• This file was extracted from the parent commit and rendered in the before run; it supplies the unchanged-behavior baseline.

▶ Before PR: true/false request without a three-distractor prompt

• Chromium submitted the true/false topic through the pre-PR dialog; the queued prompt contains the topic but no three-distractor instruction.

Before PR: queued true/false prompt and schema guidance

• The captured browser frame shows the pre-PR queued prompt beside the schema’s one-distractor true/false guidance; there is no conflict in the prompt.

▶ PR version: true/false request with conflicting distractor guidance

• Chromium submitted the same topic through the PR dialog; the queued prompt requests three distractors beside schema guidance requesting one for true/false.

PR version: queued prompt says three while schema says one

• The captured browser frame displays both instructions after submission; it makes the guidance conflict directly comparable with the before frame.

Successful Chromium execution record

• The captured command output records the working directory, exit code 0, and both observed queued prompts and schema descriptions; it confirms the asserted difference.

_{Ran code and verified through T-Rex}

[greptile-apps]

Align true/false guidance

Submitting a true/false quiz request sends an instruction to write three distractors, while the quiz tool instructs the AI to use one for true/false questions. The conflicting guidance makes the requested format less reliable. Reserve three distractors for standard multiple-choice questions. This is non-blocking.

Artifacts

Executable Chromium comparison command

• This script rendered the parent or PR dialog and submitted a topic-only request, capturing the resulting composer queue.

Rendered dialog and composer queue fixture

• This fixture mounted the real dialog and displayed its queued prompt without supplying workspace sources.

Empty-workspace browser fixture

• This page hosted the dialog with no source files, quotes, or attachments.

Parent revision submission log

• The recorded command exited 0 and captured a queued topic-only prompt without the source-material instruction.

▶ Parent revision dialog and topic-only submission

• Chromium submitted the topic-only request through the parent dialog, showing the original prompt in the queue.

Parent revision queued prompt

• The captured browser frame shows the submitted topic-only prompt without a source-material instruction.

PR revision submission log

• The recorded command exited 0 and captured the same queued request with the added source-material instruction.

▶ PR revision dialog and topic-only submission

• Chromium submitted the same request through the PR dialog, showing the changed prompt in the queue.

PR revision queued prompt

• The captured browser frame shows the added source-material instruction despite the fixture having no sources.

Chromium validation command source

• The executed script renders both dialog versions, submits the same topic, and checks the queued prompts against the tool schema; it documents the comparison.

Rendered dialog validation source

• The executed page mounts the real dialog and displays the composer queue and live schema description; it shows what the browser test exercised.

Pre-PR dialog source used for comparison

• This file was extracted from the parent commit and rendered in the before run; it supplies the unchanged-behavior baseline.

▶ Before PR: true/false request without a three-distractor prompt

• Chromium submitted the true/false topic through the pre-PR dialog; the queued prompt contains the topic but no three-distractor instruction.

Before PR: queued true/false prompt and schema guidance

• The captured browser frame shows the pre-PR queued prompt beside the schema’s one-distractor true/false guidance; there is no conflict in the prompt.

▶ PR version: true/false request with conflicting distractor guidance

• Chromium submitted the same topic through the PR dialog; the queued prompt requests three distractors beside schema guidance requesting one for true/false.

PR version: queued prompt says three while schema says one

• The captured browser frame displays both instructions after submission; it makes the guidance conflict directly comparable with the before frame.

Successful Chromium execution record

• The captured command output records the working directory, exit code 0, and both observed queued prompts and schema descriptions; it confirms the asserted difference.

_{Ran code and verified through T-Rex}