Skip to content
Documentation
Open the panel

Command lineCommands

Controlling an agent

Starting, pausing, stopping and refreshing an agent. These are the same buttons as on the toolbar of an agent in the panel.

Each command answers with a table: the state of the agent, the state the platform leads it to (wanted), the state its node reported (observed) and the actions the agent accepts now (next actions). The platform refuses an action outside that list.

1

Step 1: Start an agent

The agent starts working: its monitor looks for tasks and its executor carries them out. The project needs a Claude account in its queue.

$ kcloud agents start sandbox/release-notes

Output

FIELD         VALUE
agent         Release notes
state         starting
wanted        running
observed      missing
next actions  stop

kcloud agents check sandbox/release-notes starts a monitor cycle at once, also outside the active hours of the agent, instead of waiting for its interval, and names the number of that cycle.

2

Step 2: Pause an agent

The agent finishes its current session and starts no other. Its container keeps running, and the tasks wait in the queue.

$ kcloud agents pause sandbox/release-notes

Output

FIELD         VALUE
agent         Release notes
state         pausing
wanted        paused
observed      running
next actions  stop
3

Step 3: Resume an agent

A paused agent goes back to work.

$ kcloud agents resume sandbox/release-notes

Output

FIELD         VALUE
agent         Release notes
state         starting
wanted        running
observed      running
next actions  stop
4

Step 4: Apply the changes

Some changes of the settings of an agent need a new container. The command replaces the container and applies the changes waiting for it.

$ kcloud agents update sandbox/release-notes

Output

FIELD         VALUE
agent         Release notes
state         updating
wanted        running
observed      running
next actions  cancel
5

Step 5: Restart the container

The platform replaces the container of the agent with a new one. The memory of the agent and its tasks stay.

$ kcloud agents restart sandbox/release-notes

Output

FIELD         VALUE
agent         Release notes
state         updating
wanted        running
observed      running
next actions  cancel
6

Step 6: Cancel the operation under way

The command stops a restart or an update still under way, and a provision or a reset that failed.

$ kcloud agents cancel sandbox/release-notes --yes

Output

FIELD         VALUE
agent         Release notes
state         updating
wanted        running
observed      running
next actions
7

Step 7: Stop an agent

The agent finishes its current session, and then its container stops. With --now the container stops at once.

$ kcloud agents stop sandbox/release-notes --yes

Output

FIELD         VALUE
agent         Release notes
state         stopping
wanted        stopped
observed      running
next actions
8

Step 8: Reset the workspace

The command works on a stopped agent. The platform throws its workspace away and clones the repositories again.

$ kcloud agents reset sandbox/release-notes --yes

Output

FIELD         VALUE
agent         Release notes
state         resetting
wanted        stopped
observed      exited
next actions

Matches the panel at cloud.kikimora.dev.