The following information is not part of the official Google Antigravity documentation. It was ascertained from an exchange with the Gemini 3 agent in the Google Antigravity Code Editor on November 23rd, 2025. The behaviour described here was tested and verified on that same date, but it is subject to change.
Agent workflows are markdown files that define a series of steps for the agent to execute. They are useful for automating repetitive tasks like running the app, deploying, or running tests.
File Location
All workflow files must be located in the .agent/workflows/ directory in your workspace.
- Example:
.agent/workflows/run_app.md
File Format
The file consists of a YAML frontmatter block followed by a list of markdown steps.
Example:
---
description: A short description of what this workflow does
---
1. Step Title
command to run
// turbo
2. Another Step
another commandAnnotations
You can use special comments to control execution:
// turbo: Placed on the line immediately before a step. This tells the agent to auto-run the command in that step without asking for confirmation.// turbo-all: Placed anywhere in the file. This tells the agent to auto-run ALL commands in the workflow without confirmation.
Usage
If a workflow looks relevant to your request, or if you explicitly use a slash command (e.g., /run_app), the agent is instructed to use the view_file tool to read the workflow and execute the steps sequentially, respecting the turbo annotations
Referencing Other Workflows
There is currently no syntax for one workflow to directly “call” or reference another workflow file (like an import or function call).
Each workflow is a standalone list of steps that I read and execute.
If you need a “master” workflow that does everything, the standard approach is to create a new workflow file that combines the steps from the others into one single list.
Comments are closed