extensibility

GNU Guix 官方博客介绍如何通过 `GUIX_EXTENSIONS_PATH` 编写和分发自定义 `guix` 扩展命令，并展示了新的基于 schema 版本的扩展加载机制，该机制支持编译后的 .go 文件以显著提升性能。

A deep dive into Pi School: a free, open-source (Apache 2.0), self-paced 15-lesson course that teaches you how to use Pi by having the Pi terminal coding agent itself serve as the instructor — covering installation and setup, model boundaries, session branching, and the four-layer extension system of Skills/MCP/Code Mode/Extensions.

Omar Sansevierino argues that the next wave of coding agents and harnesses hinges on extensibility and customization, citing Claude Code mods, DeepSeek Harness, and the newly shipped Pi 1.0 with Pi Durable as evidence that builders need more malleable, domain-specific harnesses.

Anthropic introduces "mods" for Claude Code, allowing users to customize its behavior, UI, and add their own features in a few lines of TypeScript, shipped via plugins.

The author proposes a new extensible layer for Markdown that recovers 80% of its convenient syntax while allowing for custom extensions to support various use cases.

The article presents an argument for learning Lisp, highlighting its unique features like macros and extensibility that allow programmers to adapt the language to their problems, and discusses the Blub paradox to explain why programmers from less powerful languages may struggle to recognize Lisp's advantages.

F3 is a next-generation open-source data file format that uses embedded WebAssembly decoders for interoperability and extensibility, addressing limitations of legacy formats like Parquet. It is currently a research prototype from a paper published in ACM.

Rhombus v1.0, a new extensible programming language built on Racket with conventional syntax and powerful macro facilities, has been released.

Anthropic introduces 'Agent Skills' as an open standard for equipping AI agents with domain-specific expertise via composable directories of instructions and scripts. This framework enhances the portability and specialization of agents like Claude Code without requiring custom model training.