xAI released a new version of its mode called /goal inside Grok Build, the terminal coding agents. This feature targets autonomous, long-running task execution. Give the agent an implementation task that is larger, and step away.
In most coding sessions, there is a back-and forth between execution and confirmation. The agent responds to your commands, you then verify the results. /goal This loop is changed. The agent continues to perform tasks until they are completed and checked. Verification could be a matter of reviewing code or inspecting web pages, as well as executing scripts.
This mode combines this autonomy with a single command, accompanied by a progress indicator.
TL;DR
/goalGrok Build Terminal Agent can perform long autonomous tasks.- It develops a plan of action, creates a list and executes it.
- Verification is the process of examining code, inspecting webpages or executing scripts.
Status,pause,The following is a list of the most recent posts on this blog:Then,It is not clear how to do this.The live run is a way to steer.- A SuperGrok subscription or X Premium Plus is required to gain access.
What is /goal
/goal It is part of Grok Build and not an independent product. Grok Build, xAI’s coding-agent and CLI tool for software development is Grok Build. The CLI runs on your terminal. It reads files, runs commands and works on your local codebase.
Grok Build comes with several built-in agent features. The plan mode blocks editing until the plan is approved. It assigns larger jobs to parallel sub-agents. It will read you AGENTS.mdThe MCP server comes with plugins and hooks. MCP connects the agents with external data and tools.
You must have a SuperGrok subscription or X Premium Plus to access the service. Installing the CLI is as simple as:
curl -fsSL https://x.ai/cli/install.sh | bash
Sign in to your account using the username and password. /goal. It is designed to be handed off. The agent will own the work that involves multiple steps if you only describe an objective.
Why long-term execution is important
The agent receives a specific goal and not a simple prompt. It executes steps, monitors the results, and then replans if things diverge from what was planned. That is the observe–plan–act loop. This trade-off comes down to observability. Long runs result in many actions that must be monitored. /goal You can answer this question with a steering command and a progress check list.
What is the best way to get in touch with you? /goal Work
It is possible to set an objective in one line. This is done in Grok Build, not the system shell. You can use this example:
The auth module should be migrated to the API.
Build plans a strategy for the objective. The work is broken down into progress checks. After that, the program starts by completing each item on its checklist.
During a run, your account is not restricted. You are able to share additional instructions while it is working. The panel will flip to the opposite side when the goal has been achieved. You can complete the process by clicking on this link:. Each item on the checklist will then be marked as being checked.
Steps for Verification
It is important to verify. /goal The work does not stop with writing the code. The task continues to be completed until it is verified.
Verification can be done in three ways, according to xAI. Agents can review code they have produced. You can inspect web pages to check the behavior. The scripts may be executed to verify the results.
It is important to note that this applies for runs performed autonomously. The agent who only changes files can report a success rate before the actual change is made. Verification is built into the run so that the agent can test its output before finishing.
Interactive Demo
Function finish(){
if(timer){clearInterval(timer);timer=null;}
running=false;
setState(‘COMPLETE’);
statusBox.classList.add(‘complete’);
emit(“Goal complete. All checklist items verified.”, ‘ok’);
renderChecklist();
runBtn.disabled=false; gi.disabled=false;
cPause.disabled=true; cResume.disabled=true; cClear.disabled=false;
Resize();
}
Controls
cPause.onclick=function(){) return;
paused=true; setState(‘PAUSED’);
emit(“Work paused. The goal is kept.”, ‘sys’);
cPause.disabled=true; cResume.disabled=false;
};
cResume.onclick=function();
cClear.onclick=function(){
if(timer){clearInterval(timer);timer=null;}
running=false; paused=false; items=[]; idx=0;
con.innerHTML=”; setState(‘IDLE’); statusBox.classList.remove(‘complete’);
checks.innerHTML=’
No active goal
Select a target or example from the list above, and run.
‘;
setProgress();
emit(“Goal dropped. Ready for a new objective.”, ‘sys’);
runBtn.disabled=false; gi.disabled=false;
cPause.disabled=true; cResume.disabled=true; cClear.disabled=true;
gi.focus();
};
cStatus.onclick=function(){
var done=items.length?items.filter(function(i){return i.done;}).length:0;
emit(“/goal status — “+stateEl.textContent+” · “+done+”/”+items.length+” steps complete”, ‘you’);
statusBox.scrollIntoView({behavior:’smooth’,block:’nearest’});
};
runBtn.onclick=startGoal;
gi.addEventListener(‘keydown’,function(e){ if(e.key===’Enter’){ startGoal(); }});
// auto-resize srcdoc for WordPress (component width + 40pt).
Resize the image using this function(){
try{
var h = offsetHeight + 40
parent.postMessage({type:’mtp-goal-resize’, height:h}, ‘*’);
}catch(e){}
}
var ro = window.ResizeObserver ? new ResizeObserver(resize) : null;
if(ro) ro.observe(root);
window.addEventListener(‘load’,resize);
setInterval(resize, 1200);
})();

