remoteagent.online  the open-source control plane for secure, cost-controlled AI agents
Rremoteagent.online

Configure agent capabilities: tools, skills, and shell commands

Tune what an agent may do: enable tools, inject skills, allow extra shell commands, and grant file paths — all from the capabilities tab.

Updated 2026-08-22Reading time 5 min

Open the capabilities tab

  1. Go to Agents and click the gear icon on an agent.
  2. Open the Capabilities tab.
  3. You will see secure mode, skills, tools, shell commands, and file paths.

Secure mode

Toggle secure mode for a locked-down profile: read-only shell, workspace-only files, no network, browser, web, apps, notify, or skills. Nothing can loosen it via a prompt.

Skills

Skills are instruction sets injected per task. Toggle the ones the agent may use. The catalog syncs from the device, so what you see is what the client actually supports.

Tools

Toggle individual tools: shell, files, web, browser, apps, notify, and more. Disabling a tool removes it from the agent entirely.

Shell commands and paths

  1. Add extra commands one per line (mkdir, cp, mv, python3).
  2. Or use * to allow everything.
  3. Add allowed file paths outside the workspace.
  4. Save capabilities.

The local policy still wins

i
Whatever you allow here, the policy file on the device remains the final authority. A cloud setting cannot override the machine owner.