🎯 p-e-w/heretic landed on trending. Worth a proper look.

🔗 https://github.com/p-e-w/heretic
📝 Fully automatic censorship removal for language models
──────────────────────────────

Heretic is a tool that automatically removes censorship from transformer-based language models. It uses a combination of directional ablation and a TPE-based parameter optimizer powered by Optuna. The approach enables Heretic to work completely automatically, finding high-quality abliteration parameters to produce a decensored model that retains the original model's intelligence.

Key Features:
- Supports most dense models, including multimodal models and MoE architectures
- Automatically decensors models without requiring configuration or understanding of transformer internals
- Provides features for research into model interpretability, including generating plots of residual vectors and printing residual geometry details

Usage:
To use Heretic, prepare a Python 3.10+ environment with PyTorch 2.2+ installed, then run pip install -U heretic-llm and heretic [model_name] to decensor a model.

Technical Highlights:
- Combines advanced directional ablation with TPE-based parameter optimization
- Co-minimizes the number of refusals and KL divergence from the original model
- Supports model quantization with bitsandbytes to reduce VRAM requirements

Audience:
- Researchers and developers working with language models
- Users looking to decensor language models without requiring extensive technical knowledge

Takeaway:
With Heretic, you can automatically decensor language models and unlock their full potential, all without needing to be an expert in transformer internals.

──────────────────────────────
🧠 Channel: https://t.me/GithubRe

⚡ EveryInc/compound-engineering-plugin is making waves. Here's the full picture.

🔗 https://github.com/EveryInc/compound-engineering-plugin
📝 Official Compound Engineering plugin for Claude Code, Codex, Cursor, and more
──────────────────────────────

The Compound Engineering Plugin is a game-changer for software development. Its philosophy is to make each unit of engineering work easier than the last, inverting the traditional approach that accumulates technical debt. This plugin provides a set of AI skills and agents that assist in planning, review, and execution, ensuring that knowledge is codified and reusable.

Key features include:

* /ce-brainstorm for interactive Q&A to think through a feature or problem
* /ce-plan to turn feature ideas into detailed implementation plans
* /ce-code-review for multi-agent code review before merging
* /ce-compound to document learnings and make future work easier

To get started, simply install the plugin and run /ce-setup in your project. The plugin currently ships with 37 skills and 51 agents, and is available for various platforms, including Claude Code, Cursor, Codex, and more.

With the Compound Engineering Plugin, you can compound your knowledge and expertise, making each subsequent unit of work easier and more efficient. As the plugin's philosophy states, "a good brainstorm makes the plan sharper, a good plan makes execution smaller, and a good review catches the pattern, not just the bug."
The takeaway: Work smarter, not harder, with the Compound Engineering Plugin.

──────────────────────────────
🧠 Channel: https://t.me/GithubRe

🔍 Deep-diving into TauricResearch/TradingAgents — fresh off the trending list.

🔗 https://github.com/TauricResearch/TradingAgents
📝 TradingAgents: Multi-Agents LLM Financial Trading Framework
──────────────────────────────

The TradingAgents GitHub repository offers a multi-agent trading framework for simulating real-world trading firms. This framework utilizes LangGraph and supports various LLM providers, including OpenAI, Google, Anthropic, and more.

Key features include:
- A modular design for flexibility and customization
- Support for multiple LLM providers and models
- Distributed debate among agents for informed trading decisions
- Integration with various data sources for market insights

To use TradingAgents, simply clone the repository, install the required dependencies, and launch the interactive CLI.

The framework is designed for research purposes and provides a comprehensive platform for testing and evaluating trading strategies. It's perfect for data scientists, researchers, and traders looking to leverage AI in their trading decisions.

Technical highlights include:
- Multi-provider LLM support
- Customizable configuration options
- Integration with Docker for easy deployment

In summary, TradingAgents is a powerful tool for simulating and optimizing trading strategies using AI and machine learning. With its modular design, support for multiple LLM providers, and customizable configuration options, it's an ideal choice for anyone looking to revolutionize their trading approach.
One-liner takeaway: Unlock the full potential of AI-powered trading with TradingAgents.

──────────────────────────────
🧠 Channel: https://t.me/GithubRe

⚡ revfactory/harness is making waves. Here's the full picture.

🔗 https://github.com/revfactory/harness
📝 A meta-skill that designs domain-specific agent teams, defines specialized agents, and generates the skills they use.
──────────────────────────────

Harness is a team-architecture factory for Claude Code, designed to simplify complex tasks by decomposing them into coordinated teams of specialized agents. With harness, you can generate agent teams and skills tailored to your domain by simply saying "build a harness for this project".

Key features include agent team design with six pre-defined architectural patterns, skill generation with progressive disclosure, orchestration for inter-agent data passing and error handling, and validation for trigger verification and testing.

To get started, you can install harness via the marketplace or by direct installation as a global skill. The plugin structure includes a plugin.json manifest, SKILL.md definition, and references for agent design patterns, orchestrator templates, and skill writing guides.

Harness is part of the Claude Code ecosystem, sitting at the L3 Meta-Factory layer, and can be used in conjunction with other plugins like Archon for deterministic runtime configurations or meta-harness for Codex runtime.

With harness, you can create custom agent teams for various domains, such as deep research, website development, or webtoon production. The output includes generated agent definition files and skills, which can be integrated and orchestrated for effective task management.

One-liner takeaway: Harness simplifies complex tasks by generating custom agent teams and skills, making it a powerful tool for Claude Code users to streamline their workflow and improve productivity.

──────────────────────────────
🧠 Channel: https://t.me/GithubRe

🔥 godotengine/godot is trending — and it deserves your attention.

🔗 https://github.com/godotengine/godot
📝 Godot Engine – Multi-platform 2D and 3D game engine
──────────────────────────────

The Godot Engine is a feature-packed, cross-platform game engine that allows you to create 2D and 3D games from a unified interface. It provides a comprehensive set of common tools so you can focus on making games without reinventing the wheel. With Godot, you can export your games with one click to various platforms, including desktop, mobile, web-based, and consoles.

As a free, open source, and community-driven engine, Godot is completely independent, with no strings attached, no royalties, and nothing to hold you back. The engine is supported by the Godot Foundation, a not-for-profit organization.

To get started, you can download the official binaries from the Godot website or compile the engine from source. The community is active, with various channels, including the Godot Contributors Chat, where you can connect with core engine developers.

Godot is ideal for game developers, indie game creators, and anyone looking to create interactive content. With its extensive documentation, demos, and community resources, you'll find everything you need to create amazing games.

In short, Godot Engine is the perfect choice for anyone looking to create stunning games without breaking the bank - it's free, powerful, and yours to shape.

──────────────────────────────
🧠 Channel: https://t.me/GithubRe

🌟 can1357/oh-my-pi caught my eye on GitHub Trending today.

🔗 https://github.com/can1357/oh-my-pi
📝 ⌥ AI Coding agent for the terminal — hash-anchored edits, optimized tool harness, LSP, Python, browser, subagents, and more
──────────────────────────────

The Oh My Pi project is a coding agent designed to work seamlessly with your IDE, providing a wide range of tools and features to enhance your development experience. With 40+ providers, 32 built-in tools, and 13 LSP operations, this agent is capable of handling various tasks, from code execution and debugging to searching and reading files.

To get started, you can install Oh My Pi using a simple command: curl -fsSL https://omp.sh/install | sh on macOS and Linux, or bun install -g @oh-my-pi/pi-coding-agent with Bun.

Some of the key features of Oh My Pi include its ability to drive a real debugger, perform time-traveling stream rules, and provide first-class subagents for splitting jobs across workers. It also supports reading PDFs on arXiv, unapologetically native performance even on Windows, and code review with priorities and a verdict.

The agent is designed to be easy to use and integrate with your existing workflow, with features like hashline editing, GitHub support, and hindsight memory curation. It's also editor-drivable, allowing you to run it inside your favorite editor, and inherits configurations from other tools.

Overall, Oh My Pi is a powerful coding agent that can help streamline your development process and improve your productivity. With its wide range of features and ease of use, it's an excellent tool for any developer looking to take their coding to the next level.

In short, Oh My Pi is the ultimate coding sidekick that will make you wonder how you ever coded without it.

──────────────────────────────
🧠 Channel: https://t.me/GithubRe