Math-To-Manim: Instructions file for GitHub Copilot

Agent Development Guidelines

Overview

The src/agents/ directory contains the AI agent implementations that power the Math-To-Manim system. Each agent has a specific role in the pipeline from concept analysis to animation code generation.

Agent Architecture

Core Agents

1. ConceptAnalyzer: Parses user prompts and identifies core concepts, domain, and difficulty level
2. PrerequisiteExplorer: Builds reverse knowledge trees by recursively discovering prerequisites
3. MathematicalEnricher: Adds LaTeX equations and mathematical rigor to tree nodes
4. VisualDesigner: Specifies camera movements, colors, and visual metaphors
5. NarrativeComposer: Creates verbose, LaTeX-rich prompts from enriched trees
6. CodeGenerator: Translates prompts into working Manim Python code
7. VideoReviewAgent: (Planned) Automated post-render QA

Agent Communication

Agents communicate through:

• Knowledge Tree Nodes: Structured data passed between agents
• JSON Serialization: Trees are cached and reused across runs
• API Calls: Claude SDK for Claude agents, OpenAI-compatible for Kimi K2

Development Principles

Single Responsibility

Each agent should have ONE clear purpose. Don't mix concerns:

# Good - focused agent
class MathematicalEnricher:
    """Adds LaTeX equations to knowledge tree nodes."""
    def enrich_node(self, node: TreeNode) -> TreeNode:
        # Only adds mathematical content
        pass

# Bad - mixed responsibilities  
class MathAndVisualEnricher:
    """Adds both math and visual design."""
    # Violates single responsibility principle

System Prompts

Structure System Prompts Clearly:

SYSTEM_PROMPT = """
You are a {role} in the Math-To-Manim pipeline.

Your specific task:
- {task_1}
- {task_2}
- {task_3}

Input format:
{input_description}

Output format:
{output_description}

Constraints:
- {constraint_1}
- {constraint_2}

Remember: {key_principle}
"""