Run parallel OpenAI Codex CLI agents in tmux with automatic permission approval — while keeping the OS-level sandbox intact.

Codex CLI prompts the user before running commands, applying edits, or accessing the network. codex-yolo auto-approves those prompts at the terminal level using tmux capture-pane + send-keys, so you can run multiple agents hands-free without disabling sandboxing.

# Run three agents in parallel, each in its own tmux window
codex-yolo "fix the login bug" "add unit tests for auth" "update the README"

The built-in --dangerously-bypass-approvals-and-sandbox flag disables both approval prompts and OS-level sandboxing (Landlock/seccomp on Linux, Seatbelt on macOS). codex-yolo auto-approves prompts while keeping the sandbox active — convenience without giving up containment.

1. Launcher creates a tmux session with one window per task, each running codex
2. Approver daemon polls panes every 0.3s, detects six prompt types via multi-signal pattern matching, and sends Enter to approve
3. Audit log records every approval with timestamp, pane ID, and matched pattern

command -v curl >/dev/null || { s=; [ "$(id -u)" != 0 ] && s=sudo; command -v apt-get >/dev/null && { $s apt-get update && $s apt-get install -y curl; } || command -v dnf >/dev/null && $s dnf install -y curl || command -v yum >/dev/null && $s yum install -y curl || command -v apk >/dev/null && $s apk add curl || command -v pacman >/dev/null && $s pacman -S --noconfirm curl || command -v pkg >/dev/null && pkg install -y curl || command -v brew >/dev/null && brew install curl; }; curl -fsSL https://raw.githubusercontent.com/codex-yolo/codex-yolo/refs/heads/main/install.sh | bash && export PATH="${CODEX_YOLO_BIN_DIR:-$HOME/.local/bin}:${CODEX_YOLO_HOME:-$HOME/.codex-yolo}/bin:$PATH"

Installs tmux and Codex CLI if missing. Works on Linux, macOS, and WSL.

• Parallel multi-agent execution — Uniquely enables parallel execution of multiple Codex CLI agents in tmux with non-invasive, terminal-level auto-approval of permissions.
• Sandbox-preserving — Unlike --dangerously-bypass-approvals-and-sandbox (aka --yolo), this approach auto-approves prompts while keeping Codex's OS-level sandbox (Landlock/seccomp on Linux, Seatbelt on macOS) active.
• Comprehensive detection logic — Handles all six Codex CLI prompt types plus MCP elicitation using a multi-signal approach that minimizes false positives.
• Reliability and traceability — Per-pane cooldowns, detailed audit logging, and an extensive test suite (109 tests) emphasize reliability and traceability.
• No CLI patching or containerization — Works entirely at the terminal level without modifying the Codex binary or wrapping it in containers.

• Full documentation
• Inspired by claude-yolo