PRODUCTION SCENARIO
Your platform team's /release skill tags a version, pushes to the registry and posts to a Slack channel. Engineers run it by hand several times a week and must keep being able to. During routine refactors Claude has started invoking it on its own once unrelated changes are finished. The skill's steps and its tool use are correct and its output is short; the only problem is who starts it.
Which frontmatter change fits this skill?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: disable-model-invocation: true, so only a person's /release can start it
Workflows with side effects, such as deploy, commit and message-sending skills, are what this field exists for: a person can still type the command, and Claude can no longer decide to start it. Setting user-invocable to false is the right choice for background knowledge that is not an action anyone would invoke, and it would take the command away from the engineers who use it weekly. Forking the context is right when a skill's output would flood the conversation, and this one's output is short. A disallowed-tools list is right when a skill must not reach a particular tool while it runs, and its tool use here is already correct.
Official Source: https://code.claude.com/docs/en/skills ↗