Google Antigravity

Google Antigravity is the coding agent that succeeds Gemini CLI. The IDE, the CLI, and agy can read repositories, edit files, run shell commands, and call tools on a developer’s machine. TrustGuard lifecycle hooks evaluate those local actions. The hooks ship in the trustguard-gemini-cli-plugin repository, the same package Gemini CLI uses.

NeuralTrust controls

Product	Scope	Controls
TrustGuard	Evaluates tool calls before they run, and records the rest of the agent lifecycle, against an organization policy.	Block · Ask on `PreToolUse`. Telemetry for the other events

PreToolUse is the only event that can stop a tool. PostToolUse, PreInvocation, PostInvocation, and Stop are sent to TrustGuard and the hook answers {}, so the agent keeps running. An allow on PreToolUse does not suppress Antigravity’s own review prompt.

Before you start

Requirement	Notes
Egress from developer machines to `{TRUSTGUARD_BASE_URL}`	The console shows the base URL for your workspace.
The Google Antigravity collector	Agent Runtime → Collectors → Catalog → AI assistants & coding agents → Google Antigravity. Create the `tgk_…` key on its Auth tab and assign the policy on the Policies tab.
Node.js	Antigravity and the hook bootstrap both need it.
Egress to GitHub Releases	On first use the bootstrap downloads the checksum-pinned `trustguard-gemini-cli` binary.

Create the policy in Observe mode. Observe records decisions in Activity without enforcing them. Review the results, then switch the policy to Enforce. See Policies.

Developers do not need NeuralTrust accounts.

Install the hooks

git clone https://github.com/NeuralTrust/trustguard-gemini-cli-plugin.git
cd trustguard-gemini-cli-plugin
make build
python3 scripts/install-antigravity-hooks.py --user

That merges a trustguard entry into ~/.gemini/config/hooks.json and leaves every other hook in place. Antigravity reads that file at startup. Restart the IDE or CLI, then run /hooks in the CLI to confirm the commands. The user file is the default because some CLI versions ignore the workspace file .agents/hooks.json. To install it anyway:

python3 scripts/install-antigravity-hooks.py --workspace /path/to/project

macOS and Linux call trustguard/hooks/trustguard-hook.sh. Windows calls trustguard-hook.ps1. Both forward the event name to trustguard-gemini-cli.

Deploy the managed config

Antigravity and Gemini CLI share one config file. Deploy it with MDM:

{
  "data_url": "https://<your-trustguard-host>",
  "api_key": "tgk_…",
  "fail_mode": "open"
}

OS	Managed config path
macOS	`/Library/Application Support/TrustGuard/gemini-cli.json`
Linux	`/etc/trustguard/gemini-cli.json`
Windows	`%ProgramData%\TrustGuard\gemini-cli.json`

For a pilot without MDM, write the same JSON to ~/.trustguard/gemini-cli.json and chmod 600 it. fail_mode: "open" allows the tool when TrustGuard cannot be reached. fail_mode: "closed" denies PreToolUse in that case and still lets the telemetry events through, so a network blip does not freeze the agent.

What is evaluated

Antigravity event	Sent to TrustGuard	Effect on the agent
`PreToolUse` on `run_command`	The shell command (`CommandLine`)	`deny` with the TrustGuard reason, `ask`, or `allow`
`PreToolUse` on other tools	The tool name and arguments	Same
`PostToolUse`	The tool result or error	None
`PreInvocation` / `PostInvocation`	The model and invocation	None
`Stop`	Why the run ended	None

Activity shows these calls with source.application = antigravity-plugin.

Verify

Run a shell command Antigravity would execute, such as listing a directory.
In Activity, confirm an event with source.application = antigravity-plugin.
Put the policy in Enforce and add a rule that blocks that command. The next PreToolUse should return deny and Antigravity should not run it.

Catalog

AI coding agents

Agent platforms

AI gateways

Agent frameworks

Observability

Build your own

NeuralTrust controls

Before you start

Install the hooks

Deploy the managed config

What is evaluated

Verify

​NeuralTrust controls

​Before you start

​Install the hooks

​Deploy the managed config

​What is evaluated

​Verify

​Related

NeuralTrust controls

Before you start

Install the hooks

Deploy the managed config

What is evaluated

Verify

Related