Agent skills for behavior standardization

Anthropic turning Claude Skills into vendor‑neutral Agent Skills is kinda like MCP deja vu, but for agent behavior instead of tools. Think of it as: - skills-as-artifacts instead of skills-as-config - git‑versioned behaviors rather than hidden prompt spaghetti - a de facto agent package layer that IDEs and frameworks can agree on As a founder, this matters because: - You can ship agent behavior like you ship libraries - You reduce lock‑in to any one vendor’s agent format - Teams can standardize on reusable skills across products and services Is it yet another spec? Yes. Is it probably the right abstraction layer for agents long term? Possibly also yes. We had packages for code, containers for infra, MCP for tools. Agent Skills is that same energy for behavior. And if this lands, your AI features start to look a lot more like a skills registry than a pile of custom glue code.