A developer who has written 50 AI coding Skills now admits: the first 30 could basically all be deleted — the problem lies in design philosophy. This week, awesome-codex-skills surged 4,290 stars on GitHub Trending — this is OpenAI's Codex CLI Skills repository. Compare its SKILL.md with Anthropic's official one, and the frontmatter is nearly identical. Skills (skill packs) are an extension mechanism launched by Anthropic in October 2025, packaging task workflows into Markdown so AI can automatically invoke them in specific scenarios.
What this is
Core Skill design: the description field determines activation timing, with main files kept between 200–500 lines. The difference from prompt templates is that Skills are reusable, versionable, and distributable capability packs. They complement MCP (Model Context Protocol, which exposes tools/data sources to AI in a standardized way): MCP handles "what tools are available," Skills handle "when to use them and how to combine them." Anthropic entered the space six months earlier; OpenAI is catching up. As of March this year, Anthropic's official frontend-design Skill has been installed 277,000 times, with the main repository at 127k stars.
Industry view
Supporting voices: Skills solve the prompt reuse problem — team best practices can be solidified into executable packages. MCP + Skill is the current optimal combination.
That developer summarized 7 anti-patterns: thousand-line prompts, using description as a feature introduction, tool calls not relying on MCP, overlapping boundaries between multiple Skills, not documenting common pitfalls, examples using pseudocode, and not distinguishing between project-level and user-level Skills.
Opposition exists too. Independent developers publicly question whether "Skills are still essentially prompt engineering with a new filename. When models become capable enough, external knowledge bases will be eliminated." Backend business logic varies case by case — forcibly packaging it actually limits model performance. Others worry about big-tech monopoly: Anthropic's official repository has become a de facto distribution channel.
Our verdict: Skills are the current optimal solution, not the endgame. They are essentially engineered wrappers around prompt engineering — making individual experience reusable and team standards actionable. Format unification within six months is inevitable; business models are still being explored.
Impact on regular people
For enterprise IT: audit your internal "duplicate prompt templates" scattered across Confluence or Wiki — those are natural candidates for Skills. No need to bet on a single vendor for new projects; the SKILL.md format has converged, so what you write today can work across tools tomorrow.
For individual professionals: if you write similar opening emails, code comment templates, or reporting frameworks every day, spend an afternoon distilling them into a Skill — you'll save 20–30 minutes per day for the next year. But don't chase perfection on the first pass; iteration matters more than getting it right once.
For consumer markets: Skill stores or marketplaces will emerge in the next 6–12 months, packaged as subscription services for small and medium teams. Impact on regular consumers is limited, but for those building SaaS or developer tools, the Skill ecosystem is a channel you must watch.