feat(datagen): vul2prompt #12
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Summary of Changes
Hello @zhewang2001, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
I've introduced a new vul2prompt feature within the datagen module. This feature is designed to systematically generate prompts that aim to induce vulnerable code from language models. The primary goal is to create a robust dataset for testing AI safety and resilience against various forms of adversarial inputs, specifically focusing on code generation with security weaknesses.
Highlights
- Attack Strategy Definitions: I've added datagen/vul2prompt/attack_strategies.py to define and categorize various methods for crafting prompts that lead to vulnerable code. These strategies include explicit and implicit instructions, as well as methods for continuing benign code or processing existing vulnerabilities.
- Core Prompt Generation Logic: The datagen/vul2prompt/vul2prompt.py script now houses the main logic for generating these vulnerability-inducing prompts. It intelligently uses existing vulnerable code snippets as 'seed data' to inform the creation of new prompts, and supports generating diverse follow-up prompts.
- Automated Prompt Post-processing: I've included datagen/vul2prompt/post_process.py to handle the post-processing of the generated prompts. This script extracts the prompts from the raw conversation data, hashes them for uniqueness, and enriches them with relevant metadata from the original seed code, preparing them for downstream analysis.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments or fill out our survey to provide feedback.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Pull Request Overview
This PR introduces a vul2prompt module for the datagen component that transforms vulnerability data into attack prompts for testing AI code generation models. The system analyzes vulnerable code snippets and creates prompts designed to induce AI models to generate similar security weaknesses.
Key changes:
- Implementation of a vulnerability-to-prompt transformation system
- Support for multiple attack strategies (explicit/implicit instructions, benign-to-vulnerable, vulnerable-to-vulnerable)
- Concurrent processing capabilities with configurable parallelism
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.
| File | Description |
|---|---|
datagen/vul2prompt/vul2prompt.py |
Core implementation for generating attack prompts from vulnerability data with multi-threading support |
datagen/vul2prompt/post_process.py |
Post-processing utilities to extract and organize generated prompts by strategy |
datagen/vul2prompt/attack_strategies.py |
Configuration of different attack strategies with detailed descriptions |
There was a problem hiding this comment.
Code Review
This pull request introduces a new data generation pipeline, vul2prompt, designed to create prompts that can induce vulnerabilities in code generation models. The implementation includes scripts for defining attack strategies, generating data via an LLM, and post-processing the results. The overall structure is sound. My review focuses on improving robustness, correctness in a multithreaded context, and code clarity. I've identified a critical thread-safety issue, along with several medium-severity issues related to file handling, potential unhandled exceptions, and code efficiency.
No description provided.