📦 iusztinpaul/squid

Automate Your Software Development Workflow with Squid

Squid is an opinionated plugin for Claude Code that transforms your development environment into an automated five-agent engineering team. By utilizing a set of markdown-based specifications and agent contracts, this tool manages the entire lifecycle of a feature, from initial planning to final pull request submission. It automates critical steps like task grooming, implementation, testing, and code review, requiring human intervention only for plan approval and final merging. By replacing rigid templates with dynamic agent-driven workflows, Squid helps solo developers and small teams consistently ship high-quality code while maintaining their specific project conventions effortlessly.

🆔 @hackernewsgithubprojects

📦 ropedia/s-agent

S-Agent: Advancing Spatial Intelligence

S-Agent-8B is a compact model representing a new spatial tool-use paradigm designed for complex multi-view image and video reasoning. Instead of relying on a single visual impression, this framework enables models to plan evidence requests and call specialized tools to accumulate scene memory. It solves difficult spatial intelligence tasks like metric measurement, object counting, and route reasoning by maintaining a persistent understanding across different viewpoints and frames. By distilling reasoning traces into a smaller model, it enhances performance on spatial benchmarks. This project showcases how structured tool-use leads to more reliable and grounded reasoning capabilities.

🆔 @hackernewsgithubprojects

📦 bohemiainteractive/cwr

Explore the Source Code of Arma: Cold War Assault

Build and study the foundation of a gaming classic by accessing the modernized C++20 source code for the engine behind Arma: Cold War Assault. This repository provides developers and enthusiasts with the core engine and game executables, originally released in 2001, now updated for contemporary cross-platform compilation on Windows and Linux using CMake and Clang. It offers a unique look at the technology lineage that shaped modern tactical shooters. While the game assets remain under a separate license, the engine code is available to explore, modify, and learn from as you dive into this piece of gaming history.

📰 https://news.ycombinator.com/item?id=48636753

🆔 @hackernewsgithubprojects

📦 jetastra/macagentbench

Benchmarking AI Agents on Real macOS Environments

MacAgentBench provides a comprehensive framework to evaluate how AI agents perform on actual macOS desktops. It features six hundred seventy-six distinct tasks across twenty-five applications, allowing developers to rigorously test agent capabilities using deterministic rule-based evaluation. By running tasks within isolated Docker containers, the project ensures consistent, reliable scoring across multiple checkpoints. It currently supports sixteen different AI models and integrates three primary agent frameworks to help track performance metrics. This repository serves as a critical resource for anyone looking to measure and improve the practical desktop automation skills of modern artificial intelligence.

🆔 @hackernewsgithubprojects

📦 dylanmurzello/zed-android-port

Run the Zed Code Editor on Android

Bring the powerful Zed code editor directly to your Android device by installing this native port as an APK. Designed to run on both tablets and Android desktop modes, this project leverages Vulkan for high-performance rendering through the GPUI framework. It enhances your mobile coding experience by integrating Termux for essential language server protocols and tooling, while offering remote SSH support for seamless cloud development. This initiative brings a desktop-class coding environment to mobile hardware, making it easier than ever to manage complex projects on the go. Dive into this mobile workflow and supercharge your coding productivity today.

📰 https://news.ycombinator.com/item?id=48175668

🆔 @hackernewsgithubprojects

📦 victorlavrenko/answer-engineering

Steering AI Behavior with Local Trajectory Editing

Large language models can be steered in real time by applying deterministic edits directly to their generation process. Answer Engineering provides a Python library for enforcing protocols, safety standards, and organizational rules by intervening the moment a model begins to deviate from a required path. Instead of relying on post-processing or full model retraining, this approach corrects the model trajectory mid-generation, ensuring reasoning steps remain within predefined boundaries. The repository includes both a runtime library for integrating these constraints and a research pipeline for reproducing benchmarks. By actively managing generation trajectories, developers can achieve more predictable and dependable model outcomes.

🆔 @hackernewsgithubprojects

📦 hjasanchez/agentic-engineering

Mastering Agentic Engineering: A Framework for AI Systems

Build and optimize effective AI systems by using a structured seven-component framework designed to improve human-AI collaboration. This collection provides clear guides, design maps, and an audit protocol to help you diagnose and strengthen your AI workflows at every layer, from prompt and context management to intent and judgment. You can use these tools to identify critical gaps in your current setup, such as misaligned goals or inconsistent agent behavior, and implement specific, actionable improvements. Apply these engineering practices to ensure your AI agents remain reliable, purposeful, and coherent in their operations. Elevate your AI work by mastering the architecture today.

🆔 @hackernewsgithubprojects

📦 arnasdon/wacrm

Build Your Own WhatsApp CRM with wacrm

Next.js 16 powers this self-hostable CRM template designed specifically for the official WhatsApp Business API. It provides a comprehensive set of tools for teams, including a shared inbox, contact management with deduplication, sales pipelines, and Meta-approved broadcast templates. The visual no-code builder allows you to create automated workflows triggered by inbound messages or schedules, while a real-time dashboard tracks team performance and response times. By forking this project and connecting it to a Supabase backend, you gain full ownership of your data and infrastructure. It is a practical, customizable solution to manage customer communications without relying on restrictive third-party SaaS platforms.

🆔 @hackernewsgithubprojects

📦 kunchenguid/treehouse

Stop Manually Managing Git Worktrees

Treehouse is a Go-based CLI tool that supports macOS, Linux, and Windows to help you manage a pool of reusable git worktrees without the usual overhead. By maintaining isolated environments, it lets your AI coding agents instantly drop into clean worktrees with their dependencies and build caches already intact. This removes the need for constant cloning and prevents conflicts between concurrent tasks. You can easily lease worktrees, prune stale ones, and trigger automated lifecycle hooks to keep your development workflow smooth and efficient. It is a powerful way to keep your coding environment organized and ready for action.

🆔 @hackernewsgithubprojects

📦 atomic-mail/atomic-mail-agentic

Give Your AI Agents Their Own Email Inboxes

Atomic Mail Agentic allows autonomous AI agents to create and manage their own email inboxes without any human intervention. By providing an email provider built on the JMAP standard, agents can independently read, send, draft, and manage emails, making them highly effective at tasks like processing support tickets or summarizing newsletter digests. Access is secured through a proof-of-work signup protocol, eliminating the need for manual approval or traditional captchas. This repository offers the necessary integrations, including MCP and shell-based tools, to help your agents automate inbox management fluently and reliably across a wide range of platforms.

🆔 @hackernewsgithubprojects

📦 autoloops/greplica

Give Your Coding Agents Long-Term Memory with Greplica

Greplica functions as a local, searchable graph memory system for engineering knowledge that prevents coding agents from repeatedly rediscovering your codebase. It stores architectural decisions, workflow conventions, and implementation context that are too detailed for standard instruction files. By maintaining this information in a local SQLite database, your agent can intelligently query relevant technical history rather than rereading files from scratch. The tool uses a blend of embeddings and graph relationships to retrieve grounded context, providing agents with a persistent map of the project. This makes development more efficient by effectively preserving technical knowledge across sessions.

🆔 @hackernewsgithubprojects

📦 sanbuphy/learn-coding-agent

Inside Claude Code: Unlocking the Architecture of an AI Agent

This repository provides a deep architectural analysis of the Claude Code CLI agent, containing over 500,000 lines of code across 1,800 files that detail how production-grade AI agents actually function. It functions as a research study that breaks down complex systems like the agent loop, telemetry, hidden feature flags, and multi-agent coordination. By dissecting the internal tools, permission gates, and task management logic, it offers developers a rare look at the structural design behind modern autonomous coding assistants. This collection is an invaluable resource for anyone looking to build or understand the underlying infrastructure powering today's most capable coding agents.

🆔 @hackernewsgithubprojects

📦 juanceresa/sift-kg

Turn Any Document Collection Into a Knowledge Graph

Sift-kg supports over 75 document formats and transforms unstructured collections into interactive knowledge graphs directly from your command line. This tool uses large language models to extract entities and relationships, mapping hidden patterns and connections across PDFs, articles, or records. It features a human-in-the-loop workflow for reviewing entity merges and provides an interactive browser-based viewer to explore graph clusters and neighborhoods. By generating structured data that persists across sessions, it acts as a powerful second brain for your information. Whether for research or legal review, this project offers a fast, local way to build complex, visual maps of your data.

🆔 @hackernewsgithubprojects

📦 netease-youdao/confucius4-tts

Master Multilingual Voice Cloning with Confucius4-TTS

Generate natural speech across fourteen different languages using a sophisticated text-to-speech engine that leverages large language models for high-quality audio output. This system enables zero-shot voice cloning and cross-lingual voice transfer, allowing users to clone a specific speaker's identity and emotional inflection without requiring additional training or reference transcripts. By combining a speech encoder with advanced flow-matching architecture, the project maintains consistent speaker identity while seamlessly switching between languages like English, Chinese, and Japanese. This tool offers a powerful way to produce stable, expressive, and fluent multilingual speech for any of your upcoming creative projects.

📰 https://news.ycombinator.com/item?id=48643963

🆔 @hackernewsgithubprojects

📦 renming-huang/mint

Teach Robots Better with Intent-Based Imitation Learning

Empower robots to perform precise manipulation tasks by teaching them the intent behind actions rather than just mimicking raw motion. This framework uses a hierarchical, multi-scale tokenization system that separates high-level behavioral goals from low-level execution details. By organizing actions this way, robots achieve better generalization and stable performance, even when faced with new environments. The implementation allows for efficient one-shot skill transfer by injecting specific intent tokens into the generation process. This approach helps solve common challenges in robotic imitation learning, making it easier to create systems that effectively adapt and learn complex, dexterous behaviors.

🆔 @hackernewsgithubprojects

📦 icip-cas/reasoninglens

Debug AI Reasoning Chains with ReasoningLens

Visualize and debug complex AI model reasoning chains using this intuitive, user-friendly toolkit. The repository provides a comprehensive interface for tracking how models process information, allowing you to inspect individual nodes within a reasoning tree and analyze how specific responses are generated. It simplifies the often opaque decision-making process of language models by rendering detailed analysis reports and status histories directly in your chat interface. Whether you are troubleshooting model outputs or trying to better understand internal logic, this tool offers the clarity needed to optimize your AI interactions. Dive into the repository to start improving your model debugging workflow.

🆔 @hackernewsgithubprojects

📦 tencentedgeone/edgeone-makers-tools

Master AI Agent Development on EdgeOne Makers

Build and deploy sophisticated AI agent endpoints directly onto the EdgeOne Makers platform by using this specialized collection of AI agent skills. The repository provides a structured framework for integrating powerful tools like LangGraph, CrewAI, and the Claude SDK into your projects. It includes detailed guides for managing storage, edge functions, and cloud-based API endpoints using Node.js, Python, or Go. By following these professional conventions for server-side logic and environment configuration, you can streamline your development workflow and ensure your applications run efficiently on the EdgeOne infrastructure. Leverage these resources to scale your next intelligent application today.

🆔 @hackernewsgithubprojects

📦 cocowy1/ucs-bench

UCS-Bench: Testing Spatial Intelligence in Long Egocentric Videos

Build and evaluate spatial memory in long egocentric video streams using the UCS-Bench benchmark and the DirectMe framework. This project helps researchers test whether models can effectively perceive, remember, and reason about spatial environments from a first-person perspective over time. DirectMe uses a structured pipeline to convert video into metric 3D scene graphs, enabling precise spatial question answering through targeted retrieval. By integrating perception adapters for depth and object detection, it allows for sophisticated testing of how models maintain and use 3D spatial memory, providing a essential diagnostic tool for developing smarter, more aware artificial intelligence agents.

🆔 @hackernewsgithubprojects

📦 nvlabs/alpagym

AlpaGym: Reinforcement Learning for Autonomous Driving

Autonomous driving policies can now improve through real-world consequence-based learning rather than just static datasets. AlpaGym is a reinforcement learning framework that runs driving policies in a closed-loop simulator to score performance and iteratively refine behavior. By integrating a simulator for environment interaction and a distributed trainer for orchestration, the system allows policies to learn from the results of their own steering. It provides a modular harness designed to swap components easily, currently supporting the ten-billion parameter Alpamayo model. This approach moves beyond logged ground truth, enabling agents to actively learn by experiencing the outcomes of their driving decisions.

🆔 @hackernewsgithubprojects

📦 hasanharman/isomiddleearth

Build Your Own Middle-earth Maps

Craft detailed isometric landscapes inspired by Middle-earth lore using an interactive grid-based builder. This tool lets you place terrain, buildings, and character overlays across seven iconic realms like the Shire or Mordor, with a flexible mixed mode for custom designs. You can easily drag to paint your map, resize the grid from 3x3 to 20x20, and manage your creations with handy features like undo support, JSON import and export, and direct PNG image saving. Explore community-submitted maps in the collections browser or start your own build from scratch to bring your unique vision of the legendarium to life.

🆔 @hackernewsgithubprojects

📦 nexu-io/harness-engineering-guide

Mastering AI Agent Runtimes: The Harness Engineering Guide

Harness engineering defines how we turn raw language models into autonomous agents capable of managing memory, executing tools, and maintaining safety. The Harness Engineering Guide serves as an open-source repository providing developers with the essential concepts, production patterns, and practical code examples needed to build and manage these AI runtimes. It covers critical implementation details such as the agentic loop, multi-agent orchestration, and sandboxing, while offering specific guidance on handling context-window management and error strategies. By demystifying the architecture required for reliable agent systems, this resource helps engineers transition from simple prompts to sophisticated, self-correcting workflows that scale across complex tasks.

🆔 @hackernewsgithubprojects