Command reference Β· Antigravity CLI
agy --dangerously-skip-permissions
Bypasses permission restrictions at startup (launch flag).
agy --dangerously-skip-permissions is a startup flag of the Antigravity CLI. It's documented not only on the older "Using AGY CLI" docs page but also officially in the --help text of the currently shipped version (1.1.7): it automatically approves all tool-permission requests without asking. Without the flag, Antigravity asks by default before write file operations, shell commands, and web access (default setting toolPermission: request-review); the flag disables these prompts entirely for the whole session.
According to the docs, the override only applies to the running session, not permanently: the settings panel (/settings) shows that and by what a value was overridden via a flag, the persistent settings.json remains untouched and takes effect again after a restart without the flag.
The name deliberately makes the intent clear: every action proposed by the agent runs without any oversight - including potentially destructive shell commands or file deletions. The flag is suited exclusively for fully isolated, disposable environments without access to sensitive data or production systems.
π THE SAME TASK (βAUTONOMY MODE / RUN WITHOUT CONFIRMATIONβ) ON OTHER PLATFORMS
- Claude Code: /permissions
- Codex CLI: /permissions
- Cursor CLI: /run-everything
β WHEN TO USE IT?
Work in a fully isolated, disposable environment without prompts
Start agy --dangerously-skip-permissions in a Docker container or a CI sandbox that has no access to production data
Short-lived, automated test runs with no way to interact
Use agy --dangerously-skip-permissions in a VM or snapshot that's discarded after the run
β WHEN NOT TO?
Working on a normal project with sensitive data or production access
According to the --help text, the flag automatically approves all tool calls, including write file operations and shell commands, without any prompt.
Better: Keep the default setting toolPermission: request-review or selectively approve individual rules via /permissions
An agent should run unattended over a longer period
Without prompts, a single faulty suggestion can execute destructive commands unchecked before anyone can intervene.
Better: Use toolPermission: proceed-in-sandbox so risky actions at least run within the sandbox
SOURCES
- Antigravity CLI docs: Using the CLI β antigravity.google