Require evaluate in Gemini and Antigravity skills

Antigravity and similar assistants load skills — folders of instructions and helper scripts — for specialized workflows. If those skills can touch paid platforms (cloud consoles, package registries, domain registrars), require evaluate before side effects.

Define the skill

Create skills/action-governance/SKILL.md:

---
name: action-governance
description: Evaluate and audit sensitive agent tool calls
---

Ensure sensitive tool calls evaluate through Limetry (CLI, MCP, or SDK) before side effects.

Put that requirement in the skill instructions so the assistant evaluates before it writes or runs execution scripts.

See Antigravity customization for local workspace skill layout.