ctf-ai-ml

Generate adversarial examples and attack ML models in CTF challenges.

Updated Apr 30, 2026
One-click install
npx skills add https://github.com/DexterJie/codex-skills --skill ctf-ai-ml-dexterjie
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ctf-ai-ml
Source: https://github.com/DexterJie/codex-skills/tree/main/ctf-ai-ml
Command: npx skills add https://github.com/DexterJie/codex-skills --skill ctf-ai-ml-dexterjie

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires torch, transformers, numpy, scipy, Pillow, safetensors, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill provides a comprehensive toolkit for tackling AI and ML challenges in Capture The Flag (CTF) competitions, offering techniques for attacking ML models, crafting adversarial examples, and solving AI-related puzzles.

Core Features & Use Cases

  • AI/ML Techniques: Offers a range of techniques for attacking ML models, including weight manipulation, model inversion, and adversarial example generation.
  • Adversarial Examples: Generates adversarial examples to fool machine learning classifiers.
  • Model Inversion: Recovers model parameters from target outputs.
  • Use Case: When faced with a challenge that involves machine learning, this skill can provide the necessary tools to understand and exploit the underlying model.

Quick Start

Execute 'ctf-ai-ml inspect-model model.pt' to inspect the structure of a provided ML model.

Frequently Asked Questions about ctf-ai-ml

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I craft adversarial examples to fool machine learning classifiers in a CTF?

To craft adversarial examples, you can use techniques that manipulate inputs to fool machine learning classifiers. This skill provides generation methods to exploit underlying model vulnerabilities and bypass target AI security mechanisms.

What is model inversion and how does it work for recovering ML parameters?

Model inversion is a technique to recover model parameters from target outputs. This skill provides tooling to reconstruct or extract underlying machine learning model weights and architecture from observed prediction data.

Can I extract LoRA adapters and inspect safetensors models for AI CTF puzzles?

Yes, you can inspect safetensors models and exploit LoRA adapters. This skill requires Python 3 with torch and transformers to analyze model structures and solve AI-related puzzles involving weight manipulation.

Do I need internet access and Python dependencies like torch and numpy to attack ML models?

Yes, attacking ML models requires internet access for tool installation and Python 3 dependencies including torch, transformers, numpy, scipy, Pillow, and safetensors to execute the provided machine learning exploitation scripts.

What is the best way to perform membership inference attacks on machine learning models?

The best way to perform membership inference attacks is by analyzing target model prediction confidence. This skill provides techniques to determine if specific data records were included in a model's training dataset.

How do I inspect a provided ML model structure to start finding vulnerabilities?

You can inspect a provided ML model structure by executing 'ctf-ai-ml inspect-model model.pt'. This allows you to analyze the architecture and identify potential exploitation vectors for AI-related challenges.