OpenAI Codex Agentic CLI Goal Feature
TL;DR
OpenAI Codex is an open-source agentic coding CLI tool that lets developers run AI-powered coding agents in their terminal. The /goal feature (enabled via a config.toml flag) introduces an autonomous looping mode where the agent iterates across multiple sessions until a high-level project goal is fully completed. Standard AI coding sessions are single-shot or short-context; the /goal command solves the problem of long-horizon software tasks by allowing the agent to persistently loop and self-direct across many sessions until a broadly defined objective is achieved, without requiring constant human re-prompting.
What it actually is
- What: OpenAI Codex is an open-source agentic coding CLI tool that lets developers run AI-powered coding agents in their terminal. The
/goalfeature (enabled via aconfig.tomlflag) introduces an autonomous looping mode where the agent iterates across multiple sessions until a high-level project goal is fully completed. - Who built it / maintained by: OpenAI
- Status: beta
- Why it matters: Standard AI coding sessions are single-shot or short-context; the
/goalcommand solves the problem of long-horizon software tasks by allowing the agent to persistently loop and self-direct across many sessions until a broadly defined objective is achieved, without requiring constant human re-prompting. - How it compares to alternatives:
- Claude Code
- GitHub Copilot Workspace
- Aider
- Cursor
- Devin
- Cline
- GitHub stars: 87,104 · License: Apache-2.0 · Archived: no
Links
Kickstarter guide
Install the Codex CLI via npm (npm install -g @openai/codex). Navigate to your home directory and open ~/.codex/config.toml, then add [features] goals = true to enable the experimental /goal feature. Restart Codex — you will see a warning that the feature is under development — then use the /goal slash command in a session to define a high-level project objective for the agent to autonomously pursue across multiple iterations.