EditFlow

EditFlow is the official repository for the OOPSLA 2026 paper: EditFlow: Benchmarking and Optimizing Code Edit Recommendation Systems via Reconstruction of Developer Flows by Chenyan Liu, Yun Lin, Jiaxin Chang, Jiawei Liu, Binhang Qi, Bo Jiang, Zhiyong Huang, Jin Song Dong.

This repository contains both the prompt auto-tuning implementation and the VS Code extension for both editing process simulation and developer flow optimization.

🚀 Demonstrations

Note

Please click the image to watch the demo video on YouTube.

Optimization

Simulation (digital twin)

📌 Core Modules

EditFlow extension includes two independent but complementary core modules:

1. Digital twin simulator

   • Purpose: Reconstruct and simulate the entire developer code editing flow based on commit history, visualize edit suggestions at each timestamp, and evaluate the effectiveness of code edit recommendation systems.
   • Backend Service: simulation_server.py (default port: 5001)
   • Configuration: VS Code Settings > Digital Twin Simulator: Server URL

2. EditFlow Optimization

   • Purpose: Optimize existing code editing assistants by maintaining the consistency of developer editing flows, providing context-aware next-edit suggestions, and supporting interactive modification/validation of suggestions.
   • Backend Service: optimization_server.py (default port: 5002)
   • Configuration: VS Code Settings > Edit Flow: Server URL

💡 Usage and Interaction

Prerequisite

Start the corresponding backend service before using each module:

• For Digital twin simulator: Run python src/simulation_server.py at project root (Default port 5001)
• For EditFlow Optimization: Run python src/optimization_server.py at project root (Default port 5002)
• For both modules: Start both backend services
• If you want to change the port, please refer to server_config.ini and remember to update the extension configuration at: Settings > Digital Twin Simulator: Server URL or Settings > Edit Flow: Server URL as well.

1. Digital twin simulator (Simulation process visualization)

• Start the Digital twin simulator backend service: Run python src/simulation_server.py at project root directory (port 5001);
• Open VS Code;
• Press cmd + shift + p (MacOS) or ctrl + shift + p (Windows/Linux) to open the command palette;
• Type Commit simulation and press enter to activate the simulation assistant;
• Enter the commit URL to be simulated, then select the System under test and Suggestion type;
• Watch the simulation process:
  • Check edit suggestions (location + content) at each timestamp;
  • Check the flow keeping status of each suggestion;
  • View the simulation process in real time;
  • The plugin pauses after each edit suggestion is simulated (wait for user confirmation to continue);
  • Click the Resume Simulation button to simulate the next edit;
  • View the Evaluation Results after simulation completes.

2. Batch simulation (Digital twin simulator)

• Set Suggestion type and System under test in src/simulation/main.py: __main__;
• Run the batch simulation command: python -m simulation.main at src directory (ensure the Digital twin simulator backend service is running first);

3. EditFlow Optimization (Optimize existing code editing assistants)

• Start the EditFlow Optimization backend service: Run python src/optimization_server.py at project root directory (port 5002);
• Open VS Code;
• Press cmd + shift + p (MacOS) or ctrl + shift + p (Windows/Linux) to open the command palette;
• Type Flow-keeper: Suggest Next Edit and press enter to activate the Optimization assistant;
• Select the Test System;
• Start editing the project after initializing the assistant;
• Click the Suggest Next Edit button and enter an edit description to get the next edit suggestion;
• View edit suggestions in the left sidebar:
  • Click any suggested location to jump to the corresponding code position (edit suggestion shown in diff view);
  • Modify the suggested content in the right panel:
    • Click Accept button to apply the suggestion;
    • Click Reject button to discard the suggestion;
  • Suggestions do not refresh automatically – continue browsing remaining suggestions until you click Suggest Next Edit again.

🕹️ Deployments

Deploy as a user

For end-users, simply follow the instructions:

1. install the extension from the VS Code Marketplace.[TBD]

2. Install Python and dependencies:

   • Python 3.10 +
   • If you are using conda, please make sure to activate the environment before running the command.

   pip install -r requirements.txt

3. Install Node.js and dependencies:

   • Node.js 18 +

   npm install -g @anthropic-ai/claude-code

4. Download the backend models via command:

   python src/systemUnderTest/CoEdPilot/download.py

5. Set configurations at .config file.

6. Start backend models via command:

   python src/simulation_server.py
   python src/optimization_server.py

Deploy as a developer

For debugging, customization purposes, please follow the instructions

1. Install Node.js dependencies in the project root directory (~/EditFlow):

   npm install

2. Open the project directory in VS Code, press F5, choose Run Extension if you are required to choose a configuration;

3. A new VS Code window (the "development host") will open with CoEdPilot extension loaded;

4. You may debug or customize the extension via the development host;

5. To pack your customized extension, make sure yarn is installed:

   npm install -g yarn
   npm install -g vsce

6. Under the project root directory:

   yarn package

   The command will generate a .vsix file under ./extension, based on package.json file.

7. For public usage, you may release it to VS Code extension market

   • Please follow the VS Code Extension Marketplace guidelines;
   • If you modify and redistribute this extension, please clearly indicate that your version is a fork or modification, and credit this project as the original.

8. For personal usage, you may open the VS Code command palette (Ctrl + Shift + P / Cmd + Shift + P), then select Extensions: Install from VSIX... and choose the .vsix file generated in the previous step.

✍️ Citation

If you find our work helpful, please consider citing our paper:

@article{liu2026editflow,
  author = {Liu, Chenyan and Lin, Yun and Chang, Jiaxin and Liu, Jiawei and Qi, Binhang and Jiang, Bo and Huang, Zhiyong and Dong, Jin Song},
  title = {EditFlow: Benchmarking and Optimizing Code Edit Recommendation Systems via Reconstruction of Developer Flows},
  journal = {Proceedings of the ACM on Programming Languages},
  volume = {10},
  number = {OOPSLA1},
  year = {2026},
  doi = {10.1145/3798249}
}

🐛 Issues

You are welcomed report any bugs or feature requests through the GitHub issues page.

Enjoy!