OpenAI · 2026-04-30 · major
OpenAI Codex CLI 0.128.0 — /goal Loop, Plugin Marketplace Install, GPT-5.5 Docs
Codex CLI ships its own Ralph-loop primitive: set a /goal and the agent keeps looping until it judges the goal complete or the token budget is exhausted. Release also expands permission profiles, plugin marketplace, and TUI controls.
Codex CLI 0.128 lands a /goal command that keeps the agent looping until the objective is reached or a token budget runs out.
Key specs
| Version | 0.128.0 |
|---|---|
| GitHub stars | 79.3k |
| Commits in release | 49 |
What is it?
A new release of OpenAI's open-source terminal coding agent. The headline addition is /goal — a slash command that gives Codex a long-horizon objective and turns each turn into a self-evaluation against that goal. The release also expands permission profiles, adds a plugin marketplace install path, ships a codex update self-updater, and fixes a long list of TUI, sandbox, and MCP issues.
How does it work?
/goal is implemented through two prompt templates — goals/continuation.md and goals/budget_limit.md — that the runtime injects at the end of each turn. Codex evaluates progress against the stored goal, decides whether to continue, and stops on success or when the configured token budget is exhausted. The lifecycle exposes /goal pause, /goal resume, and /goal clear via app-server APIs.
Why does it matter?
Ralph-style loops have been folk knowledge among power users for months — running an agent in a while-loop until the test passes. Bringing it into Codex CLI as a first-class primitive lowers the floor for long-running autonomous runs and standardizes how budgets and stop conditions are expressed.
Who is it for?
Codex CLI users, agent harness builders.
Try it
npm install -g @openai/codex@0.128.0 && codex /goal "<objective>"