How node-recipes/llm-based-autograder works
Main Components
a) Input Layer:
• GitHub Classroom submission files
• Prompt configuration files (from /prompts directory)
• File references system (for embedding additional files)
b) Core Processing Components:
• GitHub Action Runtime Environment
• Prompt Processing System
• LLM Integration Service
• File Resolution System
c) Output Layer:
• Grading Results
• GitHub Classroom Integration
Component Relationships
• How the GitHub Action receives input from GitHub Classroom
• Prompt processing pipeline:
• Prompt file loading (/prompts directory)
• File reference resolution ({{file}} syntax handling)
• LLM interaction
• Grading result flow back to GitHub Classroom
Technical Implementation Details
• TypeScript-based components (indicated by .ts files in /src)
• Configuration elements (.env, action.yml)
• Build artifacts (dist/ directory)
• External services (LLM API)
Specific Architectural Elements
Create sections showing:
• The action's initialization flow (src/main.ts → src/index.ts)
• Configuration management (.env.example, action.yml)
• Build pipeline (rollup.config.ts, tsconfig files)
• CI/CD integration (.github/workflows)
[GitHub Classroom] → [Action Trigger]
• ↓
[Input Processing]
• Submission Files
• Prompt Configuration
• File References
• ↓
[Core Processing]
• Prompt Resolution
• LLM Integration
• Grading Logic
• ↓
[Result Generation]
• Grade Calculation
• Feedback Formation
• ↓
[GitHub Classroom Integration]
Technical Stack Indicators
• TypeScript
• GitHub Actions
• LLM Provider
• Node.js runtime