A comparison assigned a WebSocket-based Kindle task to Claude Code, Codex and Cursor using the same Sonnet 5.5 configuration. All three tools passed 16 hidden checks, but completion times ranged from about 10 to 15 minutes and usage costs differed. Their harnesses handle input transformation, tool execution and caching differently, so outcomes can vary even with the model held constant. However, this was a small test on one task, and some costs were API-equivalent estimates that are difficult to compare directly with subscription fees. Tool selection should use the user’s own recurring tasks and verification criteria.