Antigravity Harness Configuration

Scion harness configuration for the Antigravity (AGY) CLI agent.

Repository Layout

This repository is organized as follows:

• antigravity/: Contains the Scion harness configuration files.
• Dockerfile & cloudbuild.yaml: Container image build definition (in project root).
• skills/: Standard Scion SKILL.md files (in antigravity/skills/).
• .antigravitycli/: Antigravity project configuration.

Contents

File	Location	Purpose
config.yaml	antigravity/	Harness configuration (command, auth, capabilities, MCP mapping)
provision.py	antigravity/	Container-side provisioner (keyring init, onboarding bypass, MCP, hooks)
dialect.yaml	antigravity/	Hook event translation mappings (AGY events → scion lifecycle)
skills/	antigravity/	Standard Scion SKILL.md files (slash commands for AGY)
Dockerfile	Root	Container image build definition
cloudbuild.yaml	Root	Cloud Build configuration for image builds

Installing on a Scion Hub

Install the harness configuration from the antigravity/ directory:

# From the project root:
scion hc install antigravity/

The harness config bundle must include at minimum: config.yaml and provision.py. The skills/ directory and dialect.yaml are optional but recommended.

Container Image

The harness requires a container image with the AGY binary and keyring packages. The base image chain is:

core-base (dbus-x11, gnome-keyring, libsecret-1-0, libsecret-tools)
  └── scion-base (scion user, sciontool)
      └── antigravity (AGY binary at /usr/local/bin/agy)

Build and push the image:

gcloud builds submit --config=cloudbuild.yaml .

Authentication

Three auth modes are supported:

No auth (interactive login)

Launch without any token. AGY prompts for browser-based OAuth login.

scion start my-agent --harness antigravity

After logging in, export the token for future use:

# Inside the agent container:
secret-tool lookup service gemini username antigravity

Personal OAuth (oauth-token)

Set the AGY_KEYRING_TOKEN secret with the exported OAuth token JSON. Autodetected when the secret is present.

# In scion-agent.yaml:
secrets:
  - key: AGY_KEYRING_TOKEN
    description: "AGY OAuth token JSON (from secret-tool lookup)"

Enterprise/GCP (vertex-ai)

For GCP Vertex AI mode, provide the token plus GCP project settings. Trigger via explicit auth type or the AGY_USE_GCP env var.

# Explicit:
scion start my-agent --harness antigravity --harness-auth vertex-ai

# Or via env var:
# Set AGY_USE_GCP=true in agent config

Required environment variables for GCP mode:

• AGY_KEYRING_TOKEN — OAuth token JSON
• GOOGLE_CLOUD_PROJECT — GCP project ID
• GOOGLE_CLOUD_LOCATION — GCP region (defaults to global)

MCP Server Configuration

MCP servers defined in scion-agent.yaml are automatically provisioned into AGY's native config at ~/.gemini/config/mcp_config.json:

# In scion-agent.yaml:
mcp_servers:
  my-server:
    transport: stdio
    command: my-mcp-server
    args: ["--port", "8080"]
    env:
      API_KEY: "${MY_API_KEY}"

Skills

19 standard Scion skills are included in the skills/ directory and automatically staged into ~/.gemini/antigravity-cli/skills/ where AGY discovers them as slash commands.

Additional skills can be added:

• Per-template: Place in the template's skills/ directory
• Per-project: Place in /workspace/.agents/skills/{name}/SKILL.md

Resume Support

scion resume passes the -continue flag to AGY, resuming the previous conversation.