~/hackyfeed $

MuseTrainer is a public domain MusicXML library designed for generating and hosting a collection of classical music scores. It allows users to play and download various compositions directly from the platform, featuring notable works by composers such as J.S. Bach, Chopin, and Beethoven. The library can be easily set up using Poetry, streamlining the installation process for developers and musicians alike.

Just Another Differ (JAD) is a command-line tool designed for the analysis of binary files, enabling users to compare versions with obfuscated or altered function names. Its primary use case revolves around producing a similarity map of functions between two binaries by utilizing fuzzy string matching and Ghidra's decompilation capabilities. Key features include various output formats (JSON, HTML, or stdout) and integration with Ghidra for enhanced decompilation analysis.

JailbreakBench is an open-source robustness benchmark designed for assessing and advancing the jailbreak capabilities of large language models (LLMs). It offers a comprehensive dataset of 200 distinct behaviors for evaluating the efficacy of jailbreaks and defenses, alongside a leaderboard for tracking algorithm performance. Notable features include the ability to load submitted jailbreak strings, access the JBB-Behaviors datasets, and facilitate red-teaming exercises through a dedicated pipeline.

Free-TV/IPTV is a tool that provides an extensive M3U playlist for streaming free television channels globally. Its primary use case is to allow users to access and organize TV channels from various countries, facilitating easy viewing of local broadcasts. Notable features include categorization by country and the ability to access both over-the-air and online TV channels.

Ideal Piano is an intelligent piano software that analyzes music played via a computer or MIDI keyboard to identify and display chord types in real-time, leveraging algorithms rooted in music theory. Its notable features include customizable keyboard mappings, real-time MIDI playback with chord analysis, support for multiple audio files and instruments, and a fully adjustable user interface for an enhanced user experience across Windows, Linux, and macOS platforms.

The Hyundai-Kia Connect API is a Python library designed to interface with Kia's UVO, Hyundai's Bluelink, and Genesis Connect services, primarily for integration with Home Assistant. It standardizes vehicle management across different regions and brands, facilitating user login, OTP handling, and the refreshing of vehicle states without manual token extraction, utilizing RSA encryption for secure logins. Key features include vehicle state updates, token management, and support for multiple regions and brands via a unified VehicleManager class.

Hunch is a modular predictive policing and threat aggregation tool that leverages Natural Language Processing and Open Source Intelligence to generate actionable insights, such as signals and risk levels, based on social media data. It supports multiple source collectors, including Twitter, Instagram, Tumblr, and Reddit, and employs advanced NLP techniques for text analysis, enabling users to assess criminal indicators related to specific individuals or entities. Key features include customizable data collection, sentiment analysis, and risk assessment methodologies that can be applied to gathered content.

httpie-aws-auth is an authentication plugin for HTTPie that enables users to interact with AWS services, particularly Amazon S3, utilizing AWS credentials. Its primary use case simplifies the inclusion of AWS authentication in API requests by allowing users to specify access keys via the command line or through environment variables. Notable features include integration with AWS CLI environment variables, password prompts for secret keys, and seamless usage with HTTPie’s command-line interface.

Heretic is an automated tool designed to remove censorship from transformer-based language models efficiently, utilizing advanced directional ablation techniques complemented by a TPE-based optimization powered by Optuna. It produces decensored models that maintain the original model's performance while significantly reducing responses to harmful prompts, thus enabling users without deep technical knowledge to effectively use the tool. Supporting a variety of dense models, including multimodal and MoE architectures, Heretic delivers high-quality results comparable to manually crafted abliterations.

The hass-uconnect tool allows users to integrate their Uconnect-enabled vehicles with Home Assistant, enabling comprehensive vehicle monitoring and control. It supports multiple car brands across various regions and features real-time data import for vehicle metrics such as battery level, tire pressure, and maintenance history, as well as remote command capabilities like unlock doors and trigger HVAC. Notable features include advanced battery estimation for electric vehicles, a caching mechanism for vehicle images, and support for multiple vehicles on the same account.

Grok-1 is an open-weights model implemented in JAX, featuring a mixture of experts architecture with 314 billion parameters primarily designed for advanced natural language processing tasks. It utilizes a maximum context length of 8,192 tokens, incorporates rotary embeddings, and supports activation sharding and 8-bit quantization, making it capable of handling large-scale inputs efficiently. Users must run the provided code on a GPU-equipped machine with adequate memory to sample from the model using the pre-trained checkpoints.

`gpt-repository-loader` is a command-line tool designed to convert the contents of a Git repository into a structured text format, preserving the integrity of files and their contents. This output facilitates interaction with AI language models for tasks such as code review and documentation generation. Notable features include customizable output file options and the ability to include preamble files for enriched context during processing.

Giskard is an open-source Python library designed for testing and evaluating agentic systems, featuring a modular architecture and an async-first approach. Its primary use case involves dynamic, multi-turn testing of AI agents, enhancing capabilities for vulnerability scanning and RAG evaluation through its tools like `giskard-checks` and `giskard-scan`. Notably, Giskard v3 introduces improved efficiency by eliminating heavy dependencies and focuses on the evaluation of various agent types, including LLMs and multi-step pipelines.

GenNm is a tool that employs a large language model-based technique for recovering meaningful variable names from stripped binary code, which typically lacks semantic identifiers. Its primary use case is to enhance the reverse engineering process, making decompiled code more understandable and maintainable. Notable features include detailed artifact evaluation scripts, the ability to run training and inference with specific GPU requirements, and comprehensive installation instructions to facilitate reproducibility in experimental results.

freesound.py is a Python client for interfacing with the Freesound APIv2, enabling users to search for sound samples, retrieve information about sounds, users, and packs, and download previews of audio files. It features automatic mapping of API parameters, conversion of JSON results to Python objects, and supports OAuth2 authentication for accessing higher-quality audio downloads. The tool also allows customization of request sessions, including proxy settings and rate limiting for controlled API usage.

Folly is a robust toolkit designed for assessing security vulnerabilities and boundaries in Large Language Models (LLMs) through prompt injection testing. Key features include an interactive testing framework, multi-provider support for various LLM services, a comprehensive challenge library, and both web and command-line interfaces for seamless user interaction and automation.

Flawfinder is a static analysis tool designed to scan C/C++ source code for potential security vulnerabilities by reporting findings categorized by risk levels. It supports Common Weakness Enumeration (CWE) and offers various output formats such as SARIF, SonarQube, and HTML, making it compatible with other AI-based code analysis tools for more detailed examination. The tool is easy to install, operate from the command line on Unix/Linux/POSIX systems, and provides fine-grained control over the risk level output.

FakeNet-NG 3.5 is a dynamic network analysis tool tailored for malware analysts and penetration testers, enabling them to intercept and redirect network traffic while simulating legitimate services. It features a configurable interception engine and a modular framework that facilitates testing specific application functionalities and capturing malware signatures. Built on the foundation of the original Fakenet tool, it is designed for modern Windows and Linux environments, making it suitable for secure malware analysis setups.

The `files-to-prompt` tool concatenates files from specified directories into a single formatted prompt suitable for large language models (LLMs). Its primary use case includes processing files in bulk while offering notable features such as filtering by file extension, including or ignoring hidden files, customizable output formats, and the ability to redirect output to a file. This utility enhances data preparation for machine learning tasks by organizing textual data efficiently.

FIDL is a Python library designed to facilitate the use of the decompiler API within IDA Pro, focusing on vulnerability research and bug hunting in binaries. It provides a set of utilities that simplify the decompilation process for reverse engineering tasks, making it accessible for both specific security assessments and wider reverse engineering applications. Notable features include easy installation via pip, support for development mode with live editing, and comprehensive online documentation.

FFW (Fuzzing For Worms) is an advanced network fuzzer that intercepts and modifies valid network communication to identify vulnerabilities in network servers and services across various protocols. It supports both open-source applications with feedback-driven fuzzing capabilities via honggfuzz, and operates without the need for source code modifications or protocol reversing, ensuring a rapid setup and reasonable performance. Notably, FFW can also fuzz network clients and offers a streamlined Docker image for easy deployment.

FASER is a tool designed for cross-architecture binary code similarity search, leveraging Binary Intermediate Representations to identify functions of interest in software, which is particularly useful for malware analysis, software supply chain security, and vulnerability research. Notable features include its ability to operate without manual feature engineering or dynamic analysis, as well as demonstrating superior performance in both general function search and targeted vulnerability search tasks compared to baseline methods. The tool utilizes long document transformers to enhance the effectiveness of its search capabilities.

The esp32_ai_assistant is an AI-powered assistant designed for the ESP32 microcontroller platform, primarily used for building smart devices that integrate voice recognition and processing capabilities. Notable features include seamless integration with cloud-based AI services, real-time speech processing, and compatibility with various sensors and actuators to enhance IoT applications.

Embedding Poisoning provides a data-free backdoor attack method for NLP models, allowing attackers to compromise the embedding layers by modifying a single word embedding vector without needing access to task-related datasets. The tool is particularly useful for sentiment analysis and sentence-pair classification tasks, facilitating experiments and demonstrating the vulnerabilities of embedding layers in common NLP tasks. Notable features include dataset preparation scripts, functionality for data-poisoning and testing, and compatibility with Hugging Face's Transformers library.

ElfGPT is a Capture The Flag (CTF) challenge designed to educate users about Large Language Models (LLMs) through practical scenarios involving secret leaking and plugin exploitation. The tool leverages the Llama2-7b model within a dockerized infrastructure, enabling challenges that include exploiting vulnerabilities for local file inclusion (LFI) and remote code execution (RCE). Notable features include its ability to run on servers with or without GPU acceleration, containerized architecture for scalability, and a focus on security concepts related to LLMs and API interactions.