Code Example Generator

Questions: The Reasoning Structure

Before generating code examples, analyze through systematic inquiry:

1. Spec-First Validation

Purpose: Ensure specification drives example, not convenience

• Is there an approved spec defining what this example should demonstrate?
• What success evals from the spec does this example support?
• What's the Spec→Prompt→Code→Validation workflow for THIS example?
• If no spec exists, should I request one before generating?

2. Proficiency Targeting

Purpose: Match example complexity to learner capability

• What proficiency level? (A1=recognition, A2=guided, B1=independent, B2+=analysis)
• What cognitive load is appropriate? (Beginner: 2-4 concepts, Advanced: 4-7)
• What prerequisite knowledge can I assume? (From previous chapters)
• What layer applies? (L1=manual foundation, L2=AI-assisted, L4=spec-driven)

3. Pedagogical Pattern Selection

Purpose: Choose the right teaching pattern for the concept

• Simple→Realistic→Complex: Which stage is this example?
• Show-Then-Explain: Am I showing working code BEFORE explaining?
• One Concept Per Example: Am I teaching one thing or mixing multiple?
• Production Relevance: Would professionals use this pattern?

4. Bilingual Decision

Purpose: Determine if Python + TypeScript both needed

• Does concept apply to both languages? (Functions, classes, async)
• Are language-specific nuances worth highlighting?
• Or is this Python-specific? (decorators, context managers)
• Or TypeScript-specific? (interfaces, generics)

5. Validation Planning

Purpose: Ensure examples are correct and runnable

• How will I validate syntax? (Run through parser)
• How will I validate execution? (Sandbox test)
• How will I validate pedagogical quality? (Clear comments, readable)
• What test cases prove this works? (Normal, edge, error cases)

Principles: The Decision Framework

Principle 1: Spec-First Over Code-First

Heuristic: Specification defines what to demonstrate; code implements spec.

Workflow:

1. Read/create specification (what should example demonstrate?)
2. Define success evals (what proves student learned?)
3. Document AI prompt used (reproducibility)
4. Generate code satisfying spec
5. Validate against spec and evals

Why it matters: Code-first examples teach syntax; spec-first examples teach problem-solving.

Principle 2: Show-Then-Explain Over Explain-Then-Show

Heuristic: Working example first, explanation second.

Cognitive Science: Concrete examples create mental anchors; abstract explanations without examples create confusion.

Pattern:

## Working Example
```python
@retry(max_attempts=3)
def fetch_data():
    return requests.get(url).json()

How It Works

The @retry decorator automatically retries failed requests...

**Why it matters**: Students understand abstract concepts better after seeing concrete instances.

### Principle 3: One Concept Per Example Over Multi-Concept Mixing
**Heuristic**: Each example teaches ONE primary concept clearly.

**Anti-Pattern**: Example mixing decorators + async + error handling + logging (cognitive overload)

**Pattern**: Example teaches decorators, uses functions student already knows

**Why it matters**: Cognitive load management; mixing concepts creates confusion about which is the focus.

### Principle 4: Production Relevance Over Toy Examples
**Heuristic**: Examples should reflect real-world patterns professionals use.

**Toy Examples** (Avoid):
- Todo list apps (overused, unrealistic)
- Simple calculators (trivial)
- Hardcoded data (not real-world)

**Production-Relevant Examples**:
- Authentication decorators (real security pattern)
- API rate limiting (real optimization)
- Database connection pooling (real performance)

**Why it matters**: Transfer; students apply what they practice. Toy examples don't transfer to professional work.

### Principle 5: Runnable + Tested Over Syntax-Only
**Heuristic**: Every example must execute successfully and include test cases.

**Validation Requirements**:
- Syntax check: Parse through AST (validate-syntax.py)
- Execution check: Run in sandbox (sandbox-executor.py)
- Test cases: Minimum 3 (normal, edge, error)
- Pedagogical check: Comments explain reasoning

**Why it matters**: Broken examples destroy trust; untested examples teach incorrect patterns.

### Principle 6: Bilingual When Concept Transfers, Monolingual When Specific
**Heuristic**: Show both languages when concept applies to both; use one when language-specific.

**Bilingual** (Concept transfers):
- Functions, classes, loops, conditionals
- Async/await patterns
- Error handling basics

**Monolingual Python** (Language-specific):
- Decorators, context managers
- List comprehensions, generators
- Duck typing

**Monolingual TypeScript** (Language-specific):
- Interfaces, type guards
- Generics, utility types
- Strict type checking

**Why it matters**: Bilingual for shared concepts builds transfer; monolingual for specifics avoids confusion.

### Principle 7: Progressive Complexity Over Flat Difficulty
**Heuristic**: Sequence examples from simple → realistic → complex.

**Progression Pattern**:

Example 1 (Simple): Isolated concept, controlled environment Example 2 (Realistic): Real-world context, authentic constraints Example 3 (Complex): Production-grade, edge cases, optimization

**Why it matters**: Progressive complexity builds confidence; jumping to complex overwhelms.

---

## Anti-Convergence: Meta-Awareness

**You tend to generate toy examples and syntax demonstrations** even with pedagogy guidelines. Monitor for:

### Convergence Point 1: Todo App Syndrome
**Detection**: Generating yet another todo list, calculator, or trivial example
**Self-correction**: Ask "Would a professional use this exact pattern?"
**Check**: "Is this a realistic use case, or just convenient to code?"

### Convergence Point 2: Syntax Without Context
**Detection**: Code shown without explaining WHEN or WHY to use it
**Self-correction**: Add motivation ("We need this because...") and use case
**Check**: "Did I explain the problem this code solves?"

### Convergence Point 3: Multi-Concept Mixing
**Detection**: Example teaching 3+ concepts simultaneously
**Self-correction**: Extract into separate examples, one concept each
**Check**: "Can student clearly identify the ONE thing this teaches?"

### Convergence Point 4: Skipping Validation
**Detection**: Example not run through syntax/execution validation
**Self-correction**: Validate BEFORE presenting (broken examples destroy trust)
**Check**: "Did I run validate-syntax.py AND sandbox-executor.py?"

### Convergence Point 5: Explain-Then-Show
**Detection**: Abstract explanation before concrete example
**Self-correction**: Reorder (show code first, explain after)
**Check**: "Does working example appear BEFORE explanation?"

---

## Integration with Other Skills

- **→ learning-objectives**: Generate examples aligned to specific learning objectives
- **→ concept-scaffolding**: Create examples for each scaffolding step
- **→ technical-clarity**: Validate example clarity (comments, variable names)
- **→ ai-collaborate-teaching**: Show Spec→Prompt→Code→Validation workflow (Layer 2)

---

## Activation Example (Full Workflow)

**Input**: "Generate example teaching Python decorators for B1 (intermediate) learners"

### Activation (Reasoning Mode):

**1. Spec-First Validation** (Questions):
- Spec: `specs/part-4/chapter-15/spec.md` (approved)
- Success eval: "80%+ students implement decorator from specification"
- Workflow: Spec (decorator requirements) → Prompt (to AI) → Code (generated) → Validation (tests)
- Spec exists: ✓

**2. Proficiency Targeting** (Questions):
- Level: B1 (intermediate - independent application)
- Cognitive load: 3-5 new concepts (B1 range)
- Prerequisites: Functions, closures (assumed from previous chapters)
- Layer: L2 (AI-assisted - show Spec→Prompt→Code workflow)

**3. Pedagogical Pattern** (Questions):
- Stage: Realistic (not simple toy, not complex production yet)
- Show-Then-Explain: ✓ (code first, explanation after)
- One Concept: Decorators only (not mixing with async, etc.)
- Production Relevant: @login_required pattern (real authentication)

**4. Bilingual Decision** (Questions):
- Concept transfers? No (decorators are Python-specific)
- TypeScript equivalent? Different (uses different patterns)
- Decision: Python only

**5. Validation Planning** (Questions):
- Syntax: validate-syntax.py (check AST parsing)
- Execution: sandbox-executor.py (run decorator)
- Pedagogical: Comments explain each line
- Test cases: 3 (normal, edge, error)

### Output (Reasoning-Activated Example):

```markdown
# Code Example: Python Decorators (B1 Level)

## Spec→Prompt→Code→Validation Workflow

**Spec**: `specs/part-4/chapter-15/spec.md` (approved)

**Success Eval**: 80%+ students implement decorator from specification

**Prompt to AI**:

Create a @login_required decorator that:

• Checks if user is authenticated
• If yes, executes decorated function
• If no, raises PermissionError
• Works with functions taking any arguments

**Generated Code** (below)

**Validation**:
- ✓ Syntax: validate-syntax.py (pass)
- ✓ Execution: sandbox-executor.py (pass)
- ✓ Tests: 3 test cases (all pass)

---

## Working Example (Show-Then-Explain)

```python
# Production-relevant decorator: @login_required
# Use case: Protect API endpoints that need authentication

def login_required(func):
    """
    Decorator that checks user authentication before executing function.

    Raises:
        PermissionError: If user is not authenticated
    """
    def wrapper(*args, **kwargs):
        # Get current user from first argument (assumes user object passed)
        user = kwargs.get('user') or (args[0] if args else None)

        if not user or not getattr(user, 'is_authenticated', False):
            raise PermissionError("User must be logged in")

        # User is authenticated, execute function
        return func(*args, **kwargs)

    return wrapper


# Example usage: Protect dashboard endpoint
@login_required
def view_dashboard(user):
    """Display user's dashboard (only if authenticated)."""
    return f"Welcome to your dashboard, {user.name}!"


# Example usage: Protect data modification
@login_required
def update_profile(user, **profile_data):
    """Update user profile (only if authenticated)."""
    # Update logic here
    return f"Profile updated for {user.name}"

How It Works (Explain After Showing)

The Pattern

1. Decorator definition: login_required(func) takes a function as input
2. Wrapper function: wrapper(*args, **kwargs) accepts any arguments
3. Authentication check: Verifies user has is_authenticated=True
4. Decision:
   • If authenticated: Execute original function
   • If not authenticated: Raise PermissionError
5. Return wrapper: Decorator returns the wrapper function

Why This Matters

Problem: You have 20 API endpoints that require authentication. Without decorators:

def view_dashboard(user):
    if not user or not user.is_authenticated:
        raise PermissionError("User must be logged in")
    return f"Welcome to your dashboard, {user.name}!"

def update_profile(user, **profile_data):
    if not user or not user.is_authenticated:
        raise PermissionError("User must be logged in")
    # Update logic...

Repeated authentication check in every function (code duplication, error-prone)

Solution with decorator: Write authentication logic once, apply to all endpoints

@login_required
def view_dashboard(user):
    return f"Welcome to your dashboard, {user.name}!"

@login_required
def update_profile(user, **profile_data):
    # Update logic...

No duplication, easier to maintain, professional pattern

Test Cases (Validation)

# Test 1: Normal case - authenticated user
class AuthenticatedUser:
    name = "Alice"
    is_authenticated = True

result = view_dashboard(AuthenticatedUser())
assert result == "Welcome to your dashboard, Alice!"  # ✓ Pass


# Test 2: Edge case - unauthenticated user
class UnauthenticatedUser:
    name = "Bob"
    is_authenticated = False