Refactor: Use native json_schema structured output, remove manual schema injection

src/background/core/get-content-with-history.ts

@@ -19,7 +19,6 @@ Please solve the provided question based on its type and provide the correct res
 - For choice questions, output the exact index(es) of the correct answer(s).
 - For text/numerical questions, provide the exact wording or number.
 - For essay questions, provide a highly detailed and complete response, adapting exactly to the requested 'format' (HTML vs plain text) and building upon any 'initial_text' template if supplied.
-Always output strict JSON according to the requested schema block.
 `.trim();
 
 const SYSTEM_INSTRUCTION_MESSAGE = {

src/background/core/get-response.ts

@@ -21,7 +21,8 @@ async function getChatGPTResponse(
   const timeoutControler = setTimeout(() => controller.abort(), (config.timeoutValue || 20) * 1000);
 
   // Get the content to send to chatgpt
-  // Including the instructions to the AI, the images as base64 if needed, the question and the past conversation if history is set to true
+  // Including the instructions to the AI, the images as base64 if needed,
+  // the question and the past conversation if history is set to true
   const contentHandler = await getContentWithHistory(config, questionElement, question);
 
   const client = new OpenAI({
@@ -40,17 +41,16 @@ async function getChatGPTResponse(
   const requestPayload: any = {
     model: config.model,
     messages: contentHandler.messages.map(msg => ({ ...msg })),
-    max_completion_tokens: config.maxTokens || 2000 // Maximum length of the response,
+    max_completion_tokens: config.maxTokens || 2000
   };
 
+  // Use the modern json_schema structured output when a schema is available.
+  // The model is guaranteed to return schema-valid JSON — no prompt hacks needed.
   if (targetSchema) {
     requestPayload.response_format = {
-      type: 'json_object'
+      type: 'json_schema',
+      json_schema: targetSchema
     };
-
-    if (requestPayload.messages.length > 0 && requestPayload.messages[0].role === 'system') {
-      requestPayload.messages[0].content += `\n\nYou MUST respond in JSON strictly adhering to the following schema. Do NOT wrap the JSON in markdown code blocks. Output raw JSON only.\n\n${JSON.stringify(targetSchema, null, 2)}`;
-    }
   }
 
   const req = await client.chat.completions.create(fixeO(config.model, requestPayload), {
@@ -60,17 +60,13 @@ async function getChatGPTResponse(
   clearTimeout(timeoutControler);
 
   const rawResponse = req.choices[0].message.content ?? '';
-  let structuredResponse: MoodleQuestionResponse | null = null;
 
-  if (targetSchema) {
+  let structuredResponse: MoodleQuestionResponse | null = null;
+  if (targetSchema && rawResponse) {
     try {
-      const cleanedResponse = rawResponse
-        .replace(/^```(json)?[\s\S]*?\n([\s\S]*?)```$/g, '$2')
-        .replace(/^```(json)?|```$/gm, '')
-        .trim();
-      structuredResponse = JSON.parse(cleanedResponse);
+      structuredResponse = JSON.parse(rawResponse);
     } catch (e) {
-      console.error('Failed to parse structured JSON from GPT', e);
+      console.error('Failed to parse structured JSON response', e);
     }
   }