agent-cli

What problem does it solve?

When using an AI coding agent, you often need a second opinion, adversarial review, or specialized design input from a different model. Switching sessions breaks flow and loses context. This skill enables you to delegate tasks to other AI coding CLIs—Claude Code, Codex, Gemini, OpenCode, and Cursor Agent—directly from your current session, maintaining workflow continuity through non-interactive execution, persona injection, and session resume chains.

Core Features & Use Cases

• Cross-CLI Delegation: Launch other AI coding agents non-interactively for second opinions, adversarial validation (DA), design critique, architecture review, or cross-checking without leaving your session.
• Persona Injection: Instantly assign specialized roles—Devil's Advocate, designer, or architect—by injecting system prompts into delegated agents for targeted output.
• Session Persistence: Resume previous cross-model conversations using session IDs, enabling multi-round Q&A, human-in-the-loop refinement, and progressive decision-making across different AI providers.
• Environment Validation: Automatically detect your OS (macOS, Linux, WSL) and available CLIs, with built-in self-test scripts that verify non-interactive execution, autonomous driving flags, and resume functionality with timeout protection.

Quick Start

Use the agent-cli skill to delegate the current code review to Codex in DA mode for adversarial validation and resume the session with follow-up questions.