Agentic Coding

This repository contains Matt Pocock's reusable agent skills designed for efficient AI-powered development with Claude Code, Cursor, and other coding agents. The skills address four common failure modes: misalignment between developer intent and agent output (fixed by /grill-me and /grill-with-docs), excessive agent verbosity (solved through building a shared GLOSSARY.md and domain language), poor code quality (addressed via /tdd for red-green-refactor loops and /diagnosing-bugs), and architectural degradation (tackled with /improve-codebase-architecture and /to-spec for design-conscious development). Key user-invoked skills include grill-with-docs for building project domain models and documenting architecture decisions, /triage for issue workflow management, and /implement-spec for orchestrating complex multi-ticket work. The skills emphasize engineering fundamentals, composability across different agent platforms (Claude Code, Codex, Cursor, GitHub Copilot), and are installable as plugins with auto-update capability, making them immediately relevant to solo founders and small teams using AI agents for daily development work.

I found this to be a GitHub repository for gstack, an open-source software factory and Claude Code skills collection designed for AI-assisted development. However, this content does not contain news or recent developments—it's primarily a detailed technical documentation and feature guide for gstack itself. The repository showcases Garry Tan's (YC President & CEO) personal setup and productivity framework using Claude Code with 23+ specialized skills covering planning, design, engineering review, QA, security audits, and deployment. It demonstrates advanced agentic coding practices with AGENTS.md patterns, structured workflows (office-hours → plan → build → review → test → ship), and integration with tools like Codex, Cursor, and other AI coding assistants. While this directly relates to your interests in advanced AI agent techniques, efficient AGENTS.md usage, and enterprise adoption of AI coding assistants, the content is documentation rather than news. There are no recent developments, product announcements, or breaking updates to summarize here—only the tool's current capabilities and philosophical approach to agentic coding workflows.

This GitHub repository is a comprehensive curated collection of Agent Skills for AI coding agents like Claude Code, Cursor, and GitHub Copilot. It documents the progressive disclosure architecture for skills—where metadata loads efficiently (~100 tokens), full instructions load when needed (<5k tokens), and bundled resources load as required. The repository includes official Anthropic skills for document processing, spreadsheets, PDFs, and frontend design, along with hundreds of community-built skills organized by domain: development tools, security audits, productivity, marketing, data analysis, and integrations. Key resources include a basic SKILL.md template structure, compatibility matrices showing skill paths across 15+ platforms (Claude Code uses .claude/skills/), skill marketplaces like Learn Skills and skillsmp.com, and supporting CLI tools for discovery, installation, and management. The repository also highlights best practices for creating reusable procedural knowledge through skills rather than one-time prompts, emphasizes that skills are text instructions only (code execution requires MCP servers), and provides security warnings about reviewing skills from trusted sources before deployment. This directly addresses your interest in efficient AGENTS.md usage, advanced Claude Code workflows with PRD documents, and enterprise adoption patterns for AI coding assistants through structured skill libraries.