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

Allow specific shell commands for an agent

Allowlist shell commands for an agent — one per line, or * for everything — and browse the command catalog for safe defaults.

Updated 2026-08-22Reading time 4 min

Allowlist vs everything

The safest setup lists only the commands a task needs. The convenient setup uses * to allow all. Most people land in between: a curated list plus a couple of extras.

Add commands

  1. Open the agent’s Capabilities tab.
  2. Find Extra shell commands.
  3. Add one command per line: mkdir, cp, mv, python3, open, touch.
  4. Or click "Allow all commands" for *.
  5. Save capabilities.

The command catalog

The Commands page lists safe, useful commands with a short description. Click through, then copy the names you want into the allowlist field.

How execution stays safe

Commands run argv-based, never as a raw shell string, which closes the classic injection hole. Always-blocked patterns — sudo, rm -rf /, pipe-to-shell downloads — are refused regardless of the allowlist.

Dynamic tools

For named, reusable commands you can define dynamic tools in Settings. They are pushed to devices automatically, so a whole fleet gets the same helper at once.