Developer shares lightweight method for multi-agent collaboration without API costs

A developer has published a workflow for making different AI coding agents collaborate without incurring additional API usage costs. The approach leverages existing subscription plans for tools like Claude, Codex, and Gemini while avoiding extra dependencies. The method uses CLI tools in "resume mode" to continue previous sessions, allowing one agent to invoke another while preserving conversation context. This enables workflows where Claude writes a draft, Codex reviews it, and Claude revises based on the feedback. all without manual copy-pasting between interfaces. For better visibility, an alternative approach uses tmux sessions to run agents in separate panes, providing real-time monitoring of interactions. This pattern offers tighter control and debugging capabilities compared to non-interactive CLI calls. The developer notes that while multi-agent collaboration can produce more polished output, there's a risk of generating "more polished hallucination" through extended interaction chains. The workflow is presented as a practical testing pattern rather than a universal solution, emphasizing simplicity and fast experimentation over observability. The approach aims to provide different model perspectives on the same task while working within subscription-based pricing models common to coding assistants.