How to let Claude Code scheduled tasks and /loop run skills without prompting

Starting with Claude Code 2.1.325, scheduled tasks and /loop ask before running a skill that declares allowed-tools or hooks. A task that runs while nobody is around stops at that question.

To keep running that skill unattended, add a Skill(name) rule to permissions.allow in settings.json.

{
  "permissions": {
    "allow": ["Skill(deploy)"]
  }
}

Skill(name) allows only the skill with that exact name, and Skill(name *) also allows calls with arguments. Skills that Claude Desktop syncs from claude.ai take a namespace, as in Skill(anthropic-skills:pdf).

The same version also changed "Always allow" in Desktop, IDE and SDK hosts: the option to save a rule alone appears only when saving it would actually allow that command again. A scheduled task that relied on that button can now wait for you, so write the allow rules it needs into settings.json yourself.