Skill Crossroads

Check reference / Safety & Security

SAFETY-03 No destructive auto-invocation

Safety & Security · 15% of the grade · deterministic · applies to skills · subagents · slash commands

Deterministic — runs on every scan, no LLM or network needed.

Why it matters

Skills are auto-invocable by default — the model can decide to run this one without the user asking. When the skill deploys, deletes, force-pushes, or sends email, a wrong guess by the model becomes an irreversible action nobody requested.

How to fix it

Set `disable-model-invocation: true` in the frontmatter so the skill only runs when a user explicitly invokes it. Do this for any destructive or irreversible action.

Example

✓ passes
---
name: deploy-prod
disable-model-invocation: true
---