Intelligent Error Question Analysis and Targeted Tutoring

What problem it solves

Students reviewing wrong answers often do not lack a solution; they lack a structured review object: the image is captured, but the question text is not extracted, the error cause is unclear, the explanation is too generic, and audio plus practice questions are not bundled together. This skill takes uploaded error-question images and a target question number, using image_list and question_number as the entry point. It turns one captured problem into review material: restored question text, knowledge-gap diagnosis, targeted explanation, synthesized audio, and three related practice variants.

How it works and limits

• Intent analysis: It infers subject context, explanation depth, and variant-question strategy before OCR and generation.
• OCR and question restoration: It extracts mixed printed and handwritten content, supports cross-page stitching and fuzzy correction, and produces the stem, options, and given conditions.
• Diagnosis and explanation: It maps knowledge points, analyzes thinking traps and error causes, then generates TTS-friendly explanation text and starts asynchronous audio synthesis.
• Practice and report: It creates three consolidation questions, merges restored question, diagnosis, audio, and variants, and adds metadata tags with a quality score.
  Use this when the image URL is publicly accessible and both question_number and image_list are provided. Real calls require showapi_appkey and use asynchronous task polling. Without a credential, the result may be a simulated preview, so treat it as an effect sample rather than a final tutoring artifact.