Agentic Coding

This directory provides a comprehensive catalog of reusable Agent Skills for Claude Code, Cursor, and other AI coding assistants, organized by workflow category. The content covers best practices for skill structure, design patterns, and use cases across coding, design, security, research, testing, and automation domains. Official Anthropic skills address API reference, document co-authoring, PDF/document handling, web artifacts, theming, and MCP server building. Community-maintained collections emphasize practical engineering habits through skills like andrej-karpathy-skills for reducing Claude pitfalls, ponytail for minimal code changes, and production-grade agent skills. For solo founders and SaaS builders using Claude Code daily, the directory highlights PRD-focused workflow skills (ClaudeSkills for PRDs and audits), context management patterns (caveman for shorter responses, planning-with-files for Manus-style persistent markdown), and advanced "vibe coding" through UI/design skills focused on interface polish and native-quality screens. The directory maintains star counts as popularity signals only, emphasizes reviewing skill scripts and permissions before installation, and groups patterns by coding discipline—refactoring, testing, code review, browser automation, security audits, and documentation workflows.

Remotion maintains a curated list of Agent Skills that define best practices for working with AI agents like Claude Code and Cursor on Remotion projects. These skills cover comprehensive areas including best practices, React markup, studio previews, rendering, map animations, captions, SaaS architecture, interactivity, documentation lookup, and multimedia handling. Developers can install these skills via npx or add them when creating new projects, with each skill providing focused guidance for specific workflows and use cases.

The article discusses how AI agents are transforming software development into a "dark factory" model where autonomous systems handle analysis, design, coding, review, testing, and deployment. For solo founders using Claude Code and Cursor, key takeaways include: structure requirements into clear PRDs and executable specifications before deploying agents, as agents excel at implementation but lack senior engineering judgment to challenge weak requirements. The shift moves the constraint from coding speed to requirements quality and change management discipline. Critical practices include using feature flags and staged rollouts to control deployment pace, appointing a requirements engineer to work with stakeholders on machine-executable specs, and designating an agent orchestrator (architect) to supervise output quality and hold standards. The article emphasizes that as implementation becomes faster, the real bottleneck becomes organizational capacity to learn and adapt—delivery compressed to weeks requires matching change management practices borrowed from large platforms like Google, including controlled exposure, metrics-driven decisions, and political discipline around release calendars. For bounded internal applications, one senior engineer can oversee product, development, and QA; for regulated or financial work, human validation gates remain essential before production deployment.