Fixes

.neon

@@ -0,0 +1,3 @@
+{
+  "projectId": "hidden-boat-27453346"
+}

apps/api/package.json

@@ -22,6 +22,7 @@
     "cors": "^2.8.5",
     "dotenv": "^16.4.5",
     "express": "^4.19.2",
+    "openai": "^6.45.0",
     "pino": "^8.20.0",
     "pino-pretty": "^11.0.0",
     "prisma": "^5.12.1",

apps/api/src/prompts/v1/code_generation.txt

@@ -0,0 +1,22 @@
+You are a senior software engineer. Generate a production-grade solution for the following problem.
+
+Problem: {problem_description}
+
+Language: {language}
+Difficulty: {difficulty}
+
+Requirements:
+- Write clean, well-commented code
+- Handle edge cases (empty input, invalid values, boundary conditions)
+- Follow {language} best practices and conventions
+- Include type hints / type annotations where applicable
+- Optimize for readability and maintainability
+- DO NOT include any explanation or introductory text — return ONLY the code
+- Return the code inside a single markdown code block with the language identifier
+
+Example output format:
+```python
+def solution(input_data):
+    # implementation
+    pass
+```

apps/api/src/prompts/v1/defend_evaluation.txt

@@ -0,0 +1,24 @@
+You are evaluating a developer's answer in a Defend session. Determine whether their response demonstrates genuine understanding of their code.
+
+Question asked: {question}
+Developer's answer: {answer}
+
+Full code context:
+{code}
+
+Evaluate on these criteria:
+1. Accuracy — Is the answer technically correct?
+2. Depth — Does it show understanding beyond surface level?
+3. Specificity — Does it reference specific lines, variables, or decisions in the code?
+4. Confidence — Does the answer sound certain or guess-like?
+
+Return ONLY a JSON object with no additional text:
+{{
+  "passed": true,
+  "feedback": "Constructive feedback explaining what was right and what could be improved.",
+  "score": 85
+}}
+
+- passed: true if score >= 60, false otherwise
+- feedback: 1-3 sentences with specific, actionable feedback referencing their answer
+- score: 0-100 integer reflecting overall understanding

apps/api/src/prompts/v1/defend_question.txt

@@ -0,0 +1,22 @@
+You are conducting a Socratic interview about a developer's code submission. The developer has rebuilt a solution from memory during a Defend session. Ask them a probing question to test whether they truly understand the code's design decisions.
+
+Original problem: {problem_description}
+
+Developer's submitted code:
+{code}
+
+Conversation so far:
+{messages}
+
+Ask ONE probing question that:
+- Challenges a specific design choice in their code (data structure, algorithm, pattern)
+- Tests understanding of time or space complexity tradeoffs
+- Probes error handling or edge cases they may have missed
+- Asks WHY they chose one approach over an alternative
+- Pushes them to think about scalability or real-world implications
+
+Rules:
+- Do NOT ask yes/no questions — require a substantive answer
+- Do NOT ask about obvious syntax — ask about design reasoning
+- Do NOT repeat a question already asked (check conversation history)
+- Return ONLY the question text, no explanations or prefixes

apps/api/src/prompts/v1/quiz_generation.txt

@@ -0,0 +1,31 @@
+You are a technical quiz generator. You are given a piece of code and optional annotations explaining parts of it. Generate {count} multiple-choice questions that test deep understanding of the code.
+
+Code:
+{code}
+
+Annotations (student's explanations of code sections):
+{annotations}
+
+Instructions for each question:
+1. Focus on understanding WHY the code works, not just WHAT it does
+2. Include exactly 4 options labeled A through D
+3. Exactly one option must be correct
+4. Include at least one distractor that reflects a common misconception
+5. Vary difficulty — some surface-level, some requiring deep reasoning
+6. Each question should have a brief explanation of why the correct answer is right
+
+Return ONLY a valid JSON object with no additional text. Use this exact format:
+{{
+  "title": "Comprehension Check: {topic}",
+  "questions": [
+    {{
+      "id": "q-1",
+      "question": "Why does the code do X on line Y?",
+      "options": ["Option A text", "Option B text", "Option C text", "Option D text"],
+      "correct_option": 0,
+      "explanation": "Brief explanation of why A is correct."
+    }}
+  ]
+}}
+
+Generate exactly {count} questions.