scottschindler/eve-agent-design

This project is a skill for Claude Code, the AI coding assistant, that helps someone design and build AI agents using eve, a framework from Vercel for building durable, file-based AI agents. Rather than just pointing you at eve's technical documentation, this skill walks you through a full design process so you end up with something tested and production ready instead of a rough demo. The process works in stages. First it interviews you about what the agent should do, who will talk to it, what systems it needs to connect to, and what parts of the interaction are sensitive. Next it maps out the core scenarios step by step with real example data, tracking exactly what information comes in, what the agent needs to know, every external action it takes, and where a human needs to step in and approve something. From there it figures out which pieces belong to the AI model itself, which belong in regular code, and which need a human involved. Once the design is agreed on, the skill translates it into an actual project structure, building it one piece at a time and checking each part works before adding the next. It also walks through security steps such as requiring authentication on routes, adding human approval for sensitive actions, limiting what the agent's sandboxed environment can access on the network, and setting cost limits so the agent cannot run up unexpected charges. Finally, every design scenario from the discovery phase becomes an automated test, covering normal use, failure cases, the human approval flow, and multi-user scenarios, so the agent's behavior can be checked automatically going forward. It installs through the skills command line tool and works with Claude Code, Cursor, Codex, and other compatible coding agents.