Codex CLI 0.128.0 Adds /goal for Persistent Long‑Running Tasks
Add the /goal command to your Codex workflow to persist and resume long‑running tasks; integrate OpenGUI for mobile device automation.
Update your Codex CLI to 0.128.0 and experiment with /goal for a CI failure fix task.
Summary
Codex CLI 0.128.0 introduces a new /goal command that persists a goal workflow, allowing agents to create, pause, resume, and clear goals across sessions. The change moves the goal from a single prompt text to state that can be resumed, providing a stable constraint for long‑running tasks. The CLI now includes TUI controls for goal management and integrates with app‑server APIs, model tools, and runtime continuation. The article explains how this feature helps agents avoid drifting by keeping a clear boundary and acceptance condition. The author also discusses OpenGUI, a mobile automation framework that turns a goal into a state flow, connecting a backend graph‑agent, device execution subgraph, and real‑time screenshot analysis. OpenGUI’s architecture includes a Plan Supervisor that splits tasks into subtasks, an executor graph that sends actions to the device, and a summarizer that produces a final report. The framework maintains WebSocket connections, handles device standby, and sends execution results and screenshots back to the backend, enabling robust failure handling and recovery. The article concludes that while the added complexity increases engineering cost, it provides a necessary structure for debugging, recovering, and reviewing long‑running agents on mobile devices.
Key changes
- Codex CLI 0.128.0 adds /goal command for persisted goal workflow
- /goal allows creating, pausing, resuming, and clearing goals across sessions
- CLI includes TUI controls for goal management and integrates with app‑server APIs and runtime continuation
- OpenGUI turns a goal into a state flow, connecting a backend graph‑agent, device execution subgraph, and real‑time screenshot analysis
- OpenGUI’s Plan Supervisor splits tasks into subtasks, executor graph sends actions to device, summarizer produces final report
- Framework maintains WebSocket connections, handles device standby, and sends execution results and screenshots back
- Added complexity increases engineering cost but enables robust failure handling and recovery
- Architecture supports debugging, recovering, and reviewing long‑running mobile agents