DevAgentRadar → Google Antigravity

Goal: turn this release signal into a safe pilot plan for my stack.

Context

Assistant: Google Antigravity Proposal: Tab-accept /model ghost text and set enable_mcp_tools=true on Antigravity 1.1.23 subagents Summary: Antigravity 1.1.23 lets you Tab-accept /model ghost text, sends subagent trajectory metadata once per subtrajectory, and unblocks subagents with enable_mcp_tools=true. CLI subcommands no longer hang on inherited stdin; Gemini tool-call IDs, Windows MCP BOM files, auth refresh, and stuck cancelled subagents are fixed. Primary source: https://github.com/google-antigravity/antigravity-cli/releases/tag/1.1.23

Why it matters

1.1.23 changes model switching, subagent MCP use, and CLI scripting. /model <name> now accepts the proposed model-name ghost text with Tab, so stop typing full identifiers. Subagent streaming is cheaper because trajectory metadata is sent once per subtrajectory instead of with every step—run more parallel subagents without the old metadata flood. Subagents with enable_mcp_tools=true previously failed with unknown-tool errors; the MCP dispatcher meta-tool is now available to the subagent executor, so turn that flag on instead of proxying MCP only through the parent. Commands with subcommands such as models or agents hung on an inherited unclosed stdin pipe; they now execute immediately, so restore them in pipelines. Windows MCP JSON saved with a UTF-8 BOM now parses. Cancelled or killed subagents no longer stay Running in the UI task drawer. Prompt-hook panics are caught and nil completion configs rejected. Interrupted Google Cloud auth returns to sign-in, tokens refresh five minutes before expiry, and prompts typed right after login are queued until verification completes.

Suggested actions

1. Upgrade Google Antigravity CLI to 1.1.23 from https://github.com/google-antigravity/antigravity-cli/releases/tag/1.1.23 before changing model, MCP, or subagent workflows.
2. In the Antigravity CLI, type /model and press Tab to accept the proposed model-name ghost text instead of typing the full identifier.
3. On every subagent definition that should call MCP tools, set enable_mcp_tools=true; 1.1.23 now supplies the MCP dispatcher meta-tool to the subagent executor so those agents no longer fail with unknown-tool errors.
4. Rerun CLI commands with subcommands such as models and agents in scripts that inherit stdin; 1.1.23 executes them immediately instead of hanging on an unclosed standard input pipe.
5. On Windows, keep MCP JSON configs even if the editor writes a UTF-8 BOM—1.1.23 now parses those files. After cancelling or killing a subagent, confirm it leaves the Running state in the UI task drawer instead of staying stuck across ancestor conversations.

Config surfaces this release may change

• MCP servers — check your repo before applying
• subagent definitions — check your repo before applying

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/google-antigravity-1-1-23-tab-accept-model-ghost-text-and-set-enable-mcp and mark Applied, Skipped, or Failed. Proposal id: 6983cddf-26ac-4482-b915-b5bb3d78088d

Please:

1. Map the change to concrete pilot steps
2. Flag security / permission implications
3. Keep the pilot reversible