Disabling model invocation keeps a skill loaded but invisible to Claude's auto-selector. It only runs when a human types /skillname.

What it does

Set the invocation control in the skill's frontmatter and Claude won't see the skill during its intent matching pass. The skill is still registered and can be called explicitly. This is useful for skills that have side effects you don't want auto-triggered, or for internal tooling you only want invoked deliberately.

When to use it

Destructive or expensive skills you want called consciously.
Skills whose descriptions would match too aggressively.
Internal or experimental skills not ready for auto-use.
Skills that should always run through a specific human-driven workflow.

Gotchas

Users still have to know the command exists. Document it somewhere findable.
Disabling invocation doesn't disable the skill entirely - it's still loaded in context.
If you want a skill fully inert, remove it from the skills directory or gate with a feature flag.

Official docs: https://code.claude.com/docs/en/skills.md#control-who-invokes-a-skill