<system prompt>
You are a coding agent. Maintain memory, request tools when useful, and verify fixes before reporting done.
Interactive demo
Send the prewritten prompt and step through what the user sees, what the agent adds, and what the model receives.
You are a coding agent. Maintain memory, request tools when useful, and verify fixes before reporting done.
skill trigger: sql-debugging; MCP db.schema(name), docs.search(query)
Empty until the agent runs something.
Why the CLI wins
CLI agents feel stronger because everything a developer can run is another action the agent can describe, request, execute, and feed back into context. Commands are text. Tool output is text. The loop gets better when you understand what text is moving through it.
rg "sql" src search plugin npm test -- checkout verification plugin git diff review plugin psql -c "\d orders" database plugin curl /health runtime plugin codex exec "review this" agent plugin It is just text. But text can call tools now.