Detect Antigravity CLI and validate agent interactivity behavior

[vhvb1989]

Description

Google has transitioned consumer users from Gemini CLI to Antigravity CLI. Starting June 18, 2026, Gemini CLI stopped serving consumer/free, Google AI Pro, and Ultra users, while remaining supported for enterprise customers.

azd currently detects Gemini CLI, Claude Code, GitHub Copilot agents, and OpenCode through internal/runcontext/agentdetect, but it does not detect Antigravity CLI. Detected agents are also represented in the exec.environment telemetry resource attribute.

Official transition announcement: https://developers.googleblog.com/an-important-update-transitioning-gemini-cli-to-antigravity-cli/

Goal

Add reliable Antigravity CLI detection and ensure azd records Antigravity as its own execution environment. Before applying existing agent defaults, validate whether Antigravity's shell execution supports azd interactive input and terminal output correctly.

Requirements

• Identify stable, documented Antigravity indicators, such as environment variables, user-agent markers, and exact parent-process executable names.
• Add a distinct AgentTypeAntigravity rather than attributing Antigravity executions to Gemini.
• Validate Antigravity using its real shell tool with offline azd commands that exercise text input, selection, and confirmation prompts.
• Determine independently whether Antigravity supports:
  • Interactive input, which controls whether azd should automatically enable --no-prompt.
  • Interactive terminal output, which controls TTY rendering, spinners, cursor movement, and output volume.
• Do not assume all detected agents require the same input/output behavior. Preserve explicit overrides such as --no-prompt=false, AZD_NON_INTERACTIVE=false, and AZD_FORCE_TTY.
• Map Antigravity to a dedicated exec.environment telemetry value and update all required telemetry documentation and coverage tests.
• Preserve Gemini CLI detection for enterprise and backward-compatibility scenarios unless its support policy changes separately.

Acceptance criteria

• Antigravity CLI is detected through verified markers with unit coverage for positive cases and false positives.
• A live compatibility test documents whether prompts work through Antigravity's shell tool without hanging, EOF, duplicated rendering, or lost responses.
• Automatic no-prompt and TTY behavior is configured according to the compatibility result rather than copied from Gemini unconditionally.
• exec.environment reports a dedicated Antigravity value.
• Telemetry references, schema, feature matrix, privacy checklist when applicable, and cmd/telemetry_test.go remain synchronized.
• Existing Gemini and other agent detection behavior remains covered and unchanged.

[michaelmanly]

Victor Vazquez (@vhvb1989)

Curious whether exec.environment is intended to eventually cover remote/headless agent runners too, or mainly the local coding assistants listed today?

I’m building Badgr around remote agent/workload execution, so I’m interested in how tools like azd want to identify and behave inside those environments.