How do I give my agent a task?
Open the project's side panel, click the tasks tab, type the job into the box at the top and press Enter to add it. It lands in the list as a pending task. To hand it to the agent, click the task's text to open its actions, then click ▶ execute with claude. That starts the agent if it is not already running and sends it the task. The task turns to running while the agent works, then done when it finishes.
Before you start
You need a project open in the quadrant (a folder pinned to that slot). Tasks belong to that project's side panel. If you have not opened a project yet, see How do I start working on a project?.
Steps
- In the coding workspace, find the project's side panel and click the tasks tab.
- At the top is a box that reads "New task… (↵ to add · ⇧↵ for newline)". Type your task into it.
- Press Enter to add it. If you want a new line inside the task without adding it yet, press Shift+Enter.
- The task appears in the list, marked pending. This means it is queued but nothing is running yet.
- Click the task's text to open a small row of actions beneath it.
- Click ▶ execute with claude. codus starts the agent (Claude) if it is not already running, then submits the task to it. The task turns to running.
- Watch the terminal. The agent begins working, and the task shows a live timer. When the agent marks it finished, the task becomes done and a line is added to the done tab.
Add it, or send it straight over
You do not have to run a task the moment you add it. Adding it puts it in the list as pending so you (or the agent) can come back to it. When you are ready, open its actions and choose how to send it:
- ▶ execute with claude: start Claude if needed and send the task straight away.
- ✎ load to claude: start Claude and drop the task into the terminal's input without sending, so you can add more detail first, then press Enter yourself.
If your team uses the Codex agent as well, you will also see ◆ execute with codex and ✎ load to codex, which do the same thing with that agent.
You can also tell the agent directly
The tasks tab is for queuing work and keeping a record of it. You can always skip it and type a request straight into the terminal instead. Use tasks when you want a list you can run one at a time, or when you want the finished work logged in the done tab.
How do I know it worked?
The task moves from pending to running, a timer appears on it, and the agent starts typing in the terminal. When it is finished the task shows as done, and a new line appears in the done tab describing what was done.
Common problems
"I clicked the task and nothing ran." Clicking the task text opens its actions; it does not run the task. Click ▶ execute with claude in the row that appears.
"I clicked execute and nothing seems to happen." If the agent was not already running, codus has to start it first, which takes a moment. Watch the terminal: it will open and the task will be sent once it is ready.
"The task is stuck on running." The agent is often waiting for you to answer a question in the terminal. Look at the terminal and reply using the arrow keys and Enter, not the mouse. See How do I answer the agent when it asks me something?.
"I added a task but it's gone." Tasks belong to the project folder open in that quadrant. If the quadrant is now on a different folder, you are looking at a different list. Point it back at your project.