Trending on GitHub

Anthropic has launched a curated repository of high-quality plugins for Claude Code, separating internal tools developed by the company from third-party community contributions. The directory serves as a centralized marketplace for installation via the /plugin command, streamlining access to extensibility features. Users are explicitly warned to vet plugins before use, as Anthropic does not verify the security or functionality of included MCP servers or external software.

• Official directory splits internal Anthropic plugins from third-party community contributions.
• Install extensions directly via Claude Code's /plugin install command.
• Anthropic disclaims liability for plugin security and functionality.
• Review plugin homepages to understand included MCP servers and dependencies.

GitNexus is a client-side tool that indexes git repositories into interactive knowledge graphs entirely within the browser. It supports inputs from GitHub, GitLab, Azure, local paths, or ZIP files, mapping dependencies and call chains. The engine includes a built-in Graph RAG agent designed to assist with code exploration and context understanding.

• Runs fully client-side in the browser, eliminating server infrastructure needs for indexing.
• Supports multiple source types including GitHub, GitLab, Azure, local repos, and ZIP files.
• Generates interactive graphs mapping dependencies, call chains, and execution clusters.
• Includes a Graph RAG agent for enhanced code exploration and context retrieval.

Archify is a Node.js system that allows AI coding agents to generate verifiable architecture diagrams by outputting typed JSON intermediate representations. It compiles this IR deterministically into self-contained HTML or SVG, supporting sequence, data-flow, and lifecycle visuals with motion. The tool integrates with Cursor, Claude Code, and similar agents to enable pre-merge architecture reviews through delta comparisons of validated snapshots.

• Agents output typed JSON IR, ensuring diagrams are programmatically verifiable rather than hallucinated images.
• Deterministic compilation to HTML/SVG allows crisp, interactive exports with dark/light themes and motion.
• Supports five diagram types including architecture, workflow, sequence, data-flow, and lifecycle views.
• Enables before/after/delta comparisons of system maps to track exact structural changes before merging.
• Works with Cursor, Claude Code, Codex CLI, and OpenCode for in-chat diagram generation.

Marin is an open-source platform dedicated to the research and development of foundation models, with a specific focus on training large language models. It covers the entire lifecycle including data curation, transformation, filtering, tokenization, pretraining, post-training, and evaluation. The project emphasizes open development by documenting processes, experiments, and decisions transparently.

• Provides a full-stack framework for LLM training from raw data to evaluation.
• Open-source commitment includes sharing process knowledge and experimental decisions.
• Covers critical stages like data curation, filtering, and tokenization.
• Suitable for researchers needing reproducible and documented model development workflows.