Codex CLI 0.128.0 adds /goal
OpenAI's Codex CLI now supports autonomous goal-setting via a /goal command that implements a Ralph loop pattern, allowing the agent to iteratively work toward objectives within token budgets. This represents a shift toward more self-directed code generation workflows, where models can reason about task completion rather than executing single-shot requests. The feature signals OpenAI's investment in agentic coding tools that balance autonomy with resource constraints, a key tension as LLM-powered development assistants mature.
Modelwire context
ExplainerThe Ralph loop name is a reference to a specific agentic architecture pattern (observe, plan, act, reflect) rather than a proprietary OpenAI invention, which means the real story here is that Codex CLI is now exposing a well-understood agent design to everyday developers through a simple slash command rather than requiring them to wire up the loop themselves.
Modelwire has no prior coverage to anchor this to directly. This story belongs to a broader thread around agentic coding tools, where the central question is how much autonomous decision-making should be delegated to a model mid-task versus kept under explicit human control. The token budget constraint mentioned in the summary is doing real work here: it is the mechanism that keeps the loop from running indefinitely, and it is the kind of practical guardrail that distinguishes a shipping feature from a research demo.
Watch whether third-party benchmarks on multi-step coding tasks (SWE-bench variants in particular) show measurable completion rate improvements with /goal enabled versus standard single-shot Codex CLI usage. If the gains are real, expect competing CLI tools like Aider or Continue to ship equivalent loop abstractions within two to three months.
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
MentionsOpenAI · Codex CLI · Simon Willison · Ralph loop
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. The full content lives on simonwillison.net. If you’re a publisher and want a different summarization policy for your work, see our takedown page.