← All posts

AI + Dev Digest — August 28, 2026

Domain-specialized agents, an official Claude plugin registry, and frictionless browser automation top today's GitHub trending.

The pattern across today's most-starred repositories is hard to miss: AI agents are rapidly specializing. Rather than competing on raw generality, teams are building agents that are genuinely good at one thing — doing science, producing video, managing knowledge, or automating browsers without getting in the user's way. The result is an emerging taxonomy of purpose-built agent toolkits, each optimizing for a particular domain or workflow.

anthropics/claude-plugins-official: A Curated Home for Claude Code Plugins

Anthropic published an official, Anthropic-managed directory of Claude Code plugins — a single curated registry of high-quality extensions that integrate directly into the Claude Code workflow. This signals a maturation of the Claude Code plugin ecosystem: rather than scattered community repositories, developers now have a canonical source for vetted integrations. With 34,000+ stars and nearly 300 new today, it's already one of the most-followed repositories in the Anthropic GitHub org.

github.com/anthropics/claude-plugins-official

K-Dense-AI/scientific-agent-skills: Teaching Agents to Do Science

Scientific-agent-skills ships 163 validated skill files that turn AI coding agents into domain researchers capable of running biology, chemistry, and medicine workflows. Each skill is a self-contained module — literature search, protein structure prediction, drug interaction lookup — that an agent can invoke without needing bespoke prompting. It bridges the gap between "AI that can answer science questions" and "AI that can actually run a research pipeline."

github.com/K-Dense-AI/scientific-agent-skills

Tencent/BrowserSkill: Browser Automation Without the Context Switch

BrowserSkill is a CLI and browser extension from Tencent that lets AI agents automate web tasks while the user keeps working in the same browser session. Unlike headless-browser approaches that open a separate instance — with all the re-authentication headaches that entails — BrowserSkill runs inside the user's existing browser, sharing cookies and login state. It's a practical solution to one of the more persistent friction points in AI-assisted web automation.

github.com/Tencent/BrowserSkill

DietrichGebert/ponytail: The Lazy Senior Dev Heuristic

Ponytail takes an opinionated stance on AI code generation: the best output comes from agents that reason like a seasoned developer who prefers working solutions over clever ones. The library applies a set of constraints and prompting heuristics that push agents toward minimal diffs, reuse of existing patterns, and avoidance of unnecessary abstractions — essentially encoding the principle that the best code is often the code you don't write. With 114,000+ stars, it's clearly touching a nerve.

github.com/DietrichGebert/ponytail

calesthio/OpenMontage: Agentic Video Production Goes Open Source

OpenMontage bills itself as the first open-source agentic video production system, combining 12 automated pipelines with over 700 agent skill files. The project covers the full production chain — script generation, asset sourcing, editing, and export — and is designed to run end-to-end without human intervention at each step. For developers building content automation tools, it offers a complete reference architecture rather than a prototype.

github.com/calesthio/OpenMontage