Show HN: AgentSkills – Empower Agents with professional knowledge (agentskills.so)

Anthropic’s "Agent Skills" platform (demoed as Show HN) lets developers package reusable, professional knowledge and tooling into modular “skills” that Claude agents can auto‑load to extend their capabilities. The repo includes official Anthropic skills (brand styling, p5.js algorithmic art, PDF/Word/Excel/PPT toolkits, Playwright web testing, GIF/Slack optimizers, MCP server patterns for external APIs, and more) plus community‑contributed skills for dev workflows (test‑first, race‑condition fixes, code review subagents, parallel issue dispatchers). Skills are composable (chain multiple modules for multi‑step workflows), portable across Claude Web, Claude Code, Anthropic API and the Claude Agent SDK, and can include code execution so agents actually run scripts, manipulate files, and call services. This matters because it transforms LLMs from conversational assistants into automated work agents that follow enterprise guidelines and specialized workflows—reducing engineering glue code, improving reproducibility, and enabling domain‑specific subagents with tailored tools, prompts, and models. Technical implications include an explicit skill packaging model (instructions, scripts, validators, resources), a sub‑agent architecture for specialization, and MCP guidance for safe external integrations. Practical caveats are present in the examples (copyright guidance for generative art, validators for GIF sizes), highlighting that governance, testing, and review remain essential as teams deploy autonomous agent workflows.