Gemini CLI 0.47 Ships the Antigravity Migration UX Google Needed Before the June 18 Cutoff

Gemini CLI 0.47.0 is a migration release wearing a routine changelog. That is usually where the important work hides. The day users search for “Gemini CLI June 18” or “Antigravity migration,” they do not need a brand narrative. They need the tool in front of them to say, clearly and correctly, what to do next.

Google’s v0.47.0 release, published at 01:51 UTC on June 18, lands exactly where the demand spike is: a cutoff-day cluster around Gemini CLI, Code Assist, Antigravity, and alternatives. The release updates Antigravity transition banner behavior, adds documentation and migration commands for Antigravity CLI, improves /help handling for Antigravity installation and migration questions, fixes Vertex AI model mapping, makes MCP tool discovery atomic, and avoids persisting empty resume sessions.

That sounds like UX polish until you imagine being the developer whose workflow stopped making sense before coffee. On cutoff days, banners, help text, migration commands, and tool-refresh semantics are not decoration. They are the incident response path.

A banner can be an operational control

The release shipped with only three assets, but the repository around it is huge: roughly 105,391 stars, 14,100 forks, and more than 1,300 open issues at research time. The compare from 0.46.0 to 0.47.0 shows 13 commits, 65 files changed, and 11 contributors. Small diff, large blast radius. That is a familiar pattern in developer tooling: the line of code that decides whether an urgent migration notice stays visible can matter more than the splashier feature from the previous month.

PR #27676 is the tell. It ensures transition banners containing the Antigravity keyword remain visible even after the normal display limit. The default product instinct is to suppress repeated banners because users hate nagging UI. On a migration deadline, that instinct becomes dangerous. If the message says “your path is changing,” it should not disappear because a counter decided the user had already been informed. A banner that carries a deprecation or reroute notice is part of the control plane.

PR #27765 builds on that by adding platform-aware Antigravity installation guidance, migration commands, /help interception for Antigravity questions, dynamic banner injection, and an antigravity-support skill containing setup and migration documentation. That last detail is more interesting than it looks. A static doc page assumes the user leaves the workflow, finds the right section, maps it to their operating system, and returns without missing a step. A skill can package procedure inside the agent environment: install, migrate, authenticate, verify, and troubleshoot. For coding agents, that is the right direction. The tool should help migrate itself.

Atomic MCP discovery is what prevents migration from lying

The MCP tool discovery fix belongs in the same story, not a separate “reliability” bucket. PR #27619 changes refresh behavior so network drops during MCP discovery do not wipe the existing tool list and trigger false “tool not found” failures. A successful empty discovery is still distinguishable from a failed refresh. That is conservative engineering, and it is exactly what migration periods need.

Migration is when tool state is most fragile. Auth changes. Config files move. Users copy old snippets into new folders. Servers restart. VPNs drop. A transient refresh failure that destroys a known-good tool list turns a recoverable network blip into a broken agent session. Worse, it can make users distrust the new runtime for the wrong reason. The agent did not lack the tool. The runtime forgot the tool because the refresh path treated failure as truth.

Practitioners should take the hint. If you are moving Gemini CLI workflows toward Antigravity, do not validate migration with a single “hello world” prompt. Run one MCP-backed workflow before and after. Force a refresh failure if you can. Confirm the tool list is preserved on transient errors and cleared only on genuine empty discovery. If your internal workflow depends on MCP servers for issue trackers, docs, build systems, or code search, this is not optional testing. It is the difference between a controlled migration and a haunted one.

The model-routing fixes matter because enterprise users live there

The Vertex AI model mapping fix is another small item with serious implications. Many teams using Gemini in production-ish developer workflows are not just hitting a consumer endpoint. They have Vertex-backed access, enterprise auth, policy constraints, region choices, and internal model aliases. A broken mapping can look like model unavailability, permission failure, or degraded behavior depending on where it surfaces. On a migration day, ambiguity is expensive.

The release also updates auto mode to respect backend definitions for Gemini 3.5 Flash and avoids persisting empty resume sessions. Both are easy to underrate. Auto mode is where user intent meets backend policy; if it chooses against stale assumptions, developers get surprising cost, latency, or capability behavior. Empty resume sessions are the kind of lifecycle bug that pollutes state and makes troubleshooting harder. Individually minor, together they point to the same theme: a CLI that wants to be a dependable agent surface has to manage state, routing, and recovery like infrastructure.

The funny recursion is that Gemini Code Assist itself appears in the PR review trail summarizing changes to the migration path for Google’s own agent tooling. That is not proof of quality, but it is a decent symbol of where we are: agents are now helping maintain the rails users stand on while moving between agents. The rails need to be very boring.

For engineers affected by the June 18 cluster, the checklist is practical. Upgrade to Gemini CLI 0.47.0 or later. Use in-tool /help for Antigravity installation and migration rather than stale snippets from search results. Confirm the banner text matches your platform and migration path. Re-authenticate where required. Run a Vertex-backed task if you use Vertex. Run an MCP-backed task. Run a resume-session task and make sure empty sessions are not persisted. If Gemini or Antigravity sits in a critical path, test a fallback agent before you need it: Codex, Claude Code, OpenCode, Qwen Code, or whatever your team can actually support.

The industry lesson is broader than Google. Migration UX is engineering, not copywriting. Developer tools now carry enough configuration, auth, model routing, MCP state, and session history that moving users requires executable guidance, not an announcement post. Gemini CLI 0.47.0 is valuable because it treats the migration moment as something the runtime must actively handle.

That is the right bar. If an agent runtime wants to change the path under developers, it should also ship the handrail.

Sources: Google Gemini CLI v0.47.0 release, Antigravity migration PR, transition banner PR, atomic MCP discovery PR, GitHub compare