Claude Skills vs. Commands vs. Sub-Agents (neon.com)

🤖 AI Summary
Anthropic’s Claude Skills are simple bundles — each a folder with a SKILL.md that describes a workflow in plain language (and can reference executables for tasks like image/PDF editing) — but Neon has packaged four of them into a reusable Claude Code plugin plus an MCP (Model Context Protocol) server so Claude can call Neon’s APIs at runtime. Claude keeps only a skill’s name/description in its global context and dynamically loads the full workflow when needed, which keeps sessions small and makes workflows available on-demand: ask “Integrate Neon with Drizzle” and Claude pulls the Drizzle skill, scaffolds schema/connection steps, and executes the guided workflow. This matters because skills sit between rigid custom commands and autonomous subagents — they’re template-driven and context-aware, enabling deterministic multi-step automation without spinning up a separate reasoning context. Neon’s plugin includes four skills (Drizzle ORM integration, serverless driver setup, Management API workflows for creating/provisioning databases, and a Neon best-practices knowledge skill) and an MCP bridge that gives Claude runtime access to project info, connection strings, provisioning, and validation. Installation is via the Claude Code marketplace; once added, Claude auto-detects the Neon MCP connection so its tools can directly operate on Neon resources. This pattern makes repeatable developer workflows portable, auditable, and efficient for AI-assisted engineering.
Loading comments...
loading comments...