[!NOTE] This project is based on research from the Data Intelligence Lab at the University of Hong Kong and the DeepCode repository.

Overview

OpenCode Paper2Code is a professional OpenCode plugin that transforms academic papers into executable, verifiable code implementations. It provides a structured plan→implement→verify workflow with built-in safety boundaries and comprehensive validation.

Key Features

	Feature	Description
🧠	Seven-Step Workflow	Structured process from paper ingestion to final merge recommendations
🔒	Safety-First Design	All file operations confined to isolated run directories
📝	Chinese Documentation	All generated code includes detailed Chinese comments
✅	Verification Built-in	Each implementation includes tests and validation methods
⚡	OpenCode Native	Full compatibility with OpenCode hooks, commands, and plugins
📊	Artifact Tracking	Comprehensive logging and artifact management

The Seven-Step Paper2Code Workflow

Paper2Code follows a strict seven-step sequence to ensure reliable, reproducible results:

1. Initialize Run Directory

Creates an isolated runDir with a unique runId. All subsequent operations stay within this boundary.

2. Paper Content Extraction

Ingests paper sources (URLs, PDF files, or text) and extracts raw content to the run directory.

3. Intelligent Document Segmentation

Splits the paper into semantically meaningful segments (algorithms, formulas, experiments) with unique IDs.

4. Generate Detailed Implementation Plan

Produces a comprehensive artifacts/plan.yaml with five required sections, each referencing specific segment IDs as evidence.

5. Code Generation & Testing

Generates complete implementation code and test files in the generated/ directory, with Chinese comments and consistent style.

6. Minimal Verification Execution

Runs essential checks to validate core functionality and ensure the code executes without errors.

7. Final Output & Merge Recommendations

Provides a concise report with run directory path, artifact inventory, and guidance for integrating into existing projects.

Installation

Install via npm

npm install opencode-paper2code@latest

Declare the Plugin in OpenCode Config

Add the plugin to your project-level .opencode/opencode.json or global ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-paper2code@0.1.2"]
}

Recommendation: Pin the version number to avoid unexpected changes from latest.

Usage

Command Line

opencode run --command paper2code

Conversational Form

You can invoke Paper2Code through conversation with structured inputs:

Remote Paper + Requirements

/paper2code paper url: https://arxiv.org/abs/2104.09864; requirements: implement with PyTorch, include unit tests, all comments in Chinese, provide training and inference scripts

Local Paper + Requirements

/paper2code paper file: /path/to/papers/xxx.pdf; requirements: first implement MVP, then add performance optimizations, output README and reproduction steps

Recommended Input Structure

Paper source: ...
Requirements: 1) ... 2) ... 3) ...
Constraints: ... (e.g., "do not simplify core algorithms")

Paper2Code Agent

The Paper2Code agent is a specialized subagent configured with:

• Mode: subagent
• Tools: paper2code_* (all Paper2Code-specific tools)
• Prompt: Comprehensive seven-step workflow instructions with safety constraints

When activated, it strictly follows the seven-step sequence, ensuring no step is skipped or shortened.

Safety & Constraints

• File Operations: All writes are confined to the runDir directory
• Path Escapes: Prohibited via validation of all target paths
• Input Validation: Strict schema validation for all inputs
• Artifact Integrity: Atomic writes and content hashing prevent corruption

Development

Build

bun run build

Lint & Type Check

bun run lint
bun run typecheck

Test

bun run test

License

MIT © 2025 Data Intelligence Lab, The University of Hong Kong; 2026 mostly925

Acknowledgments

University of Hong Kong Data Intelligence Lab for the foundational research DeepCode - This project is based on research from the DeepCode repository OpenCode for the extensible plugin architecture

---