A new lightweight pattern lets developers run Claude, Codex, and Gemini agents side by side without paying for API usage. By invoking models through the CLI with a resume flag, one agent can hand off a draft to another for review, then loop until the output stabilizes.

The approach sidesteps SDKs and extra dependencies, making it easy to test cross‑vendor perspectives. A draft is written in Claude, passed to Codex with the command `codex exec resume --last "prompt"`, and then revised. If deeper visibility is needed, the same workflow can run inside tmux panes, allowing real‑time monitoring of each agent’s output.

This setup suits quick experiments where speed and simplicity trump observability. It exposes the trade‑off between generating polished text and risking compounded hallucinations. For teams already subscribed to LLM services, the technique offers a cost‑effective way to orchestrate multi‑agent collaboration.

In sum, the method demonstrates that with minimal tooling, developers can harness multiple LLMs for iterative drafting, reviewing, and summarizing without incurring extra API fees.