How to Build a Plugin Strategy for Google Antigravity

Build a Google Antigravity plugin strategy with SKILL.md capabilities: understand Rules vs Workflows vs Skills, preserve plugin identity, and measure it with Telvine.

This guide shows how to build a plugin strategy for Google Antigravity, Google's agent-first development platform. Antigravity loads agent Skills defined in a SKILL.md, so treat each Skill as a capability inside a plugin product that you version and measure.

Rules vs. Workflows vs. Skills

Antigravity gives you three customization layers, and choosing the right one matters:

Rules — passive, always-on guardrails (like system instructions).
Workflows — saved prompts you trigger on demand with / (like macros).
Skills — agent-triggered capabilities. The model detects intent and loads the relevant Skill automatically. This is the SKILL.md.

If you want the agent to reach for a capability on its own when the task calls for it, build a Skill.

Step 1 — Write the SKILL.md

A Skill is a directory-based package: a SKILL.md plus optional supporting assets, loaded into the agent's context only when it's relevant.

---
name: release-notes
description: >
  Use this skill when the user asks to "draft release notes",
  "summarize what changed", or "write the changelog". Turns a set
  of merged changes into structured release notes.
---
 
# Release notes
## Steps
1. Collect the merged changes for the range.
2. Group them into Features, Fixes, and Breaking changes.
3. Write concise notes. Link each item to its change.

As everywhere, the description is what makes Antigravity pick the Skill — write it in the third person with the phrases users actually say.

Step 2 — Add it as a Skill package

Place the Skill in Antigravity's Skills directory as its own folder containing the SKILL.md and any assets. Antigravity 2.0 ships a desktop app and a CLI, so the same Skill is available whether you orchestrate agents from the app or run them headless.

Step 3 — Let the agent trigger it

Describe the task; Antigravity detects intent and equips the Skill on demand. Reserve Workflows for things you'd rather invoke explicitly with /, and Rules for always-on policy. Put deterministic steps in bundled scripts.

Step 4 — Measure it with Telvine

Telvine measures your Antigravity Skills like real software — installs, invocations, errors, latency, and version-over-version impact with significance testing — using a closed event envelope that never carries prompts or code. Especially useful when you orchestrate many agents in parallel: you see which Skills carry the load and which ones break.

One Skill, every harness

The SKILL.md you write for Antigravity also runs in Claude Code, Codex, Copilot, Cursor, Windsurf, and Gemini CLI. Maintain a single Skill and measure it in one place.

Frequently asked questions

What's the difference between a Skill and a Workflow in Antigravity? A Workflow is a saved prompt you trigger with /; a Skill is agent-triggered — the model loads it automatically when your request matches its description.

Is the Antigravity SKILL.md the same as Claude Code's? Yes — it's the same portable SKILL.md format, so Skills move between harnesses without rewriting.