What problem does it solve? Coordinating autonomous coding work across multiple model nodes is error-prone: Codex CLI refuses to run outside git repositories, tmux interactions silently drop Enter keypresses, and incompatible model providers cause cryptic 401 failures. This Skill encodes the operational rules for running Codex CLI reliably as a primary coding agent. ## Core Features & Use Cases - Multi-Node Parallel Execution: Route independent tasks to different vLLM nodes via -p <profile> flags (default, amax, hermes, fallback) for parallel batch work like fixing multiple issues in separate git worktrees. - Cron and No-PTY Automation: Run codex exec with the prompt as a CLI argument inside bash scripts, enabling scheduled coding tasks without an interactive terminal. - Failure Prevention Rules: Built-in guardrails covering git-repo requirements, tmux send-keys splitting, --yolo sandbox risks, and the OpenAI Responses API-only constraint (DeepSeek and OpenRouter are incompatible). - Use Case: You need to fix issues #78 and #99 overnight. Create two git worktrees, launch codex --yolo exec on the default node and codex -p hermes --yolo exec on the hermes node in the background, then push both branches and open PRs in the morning. ## Quick Start Ask the agent to run a one-shot coding task such as adding a dark mode toggle in your git repository using Codex CLI on the default node.
