Google Antigravity release notes: CLI 1.0.14 keeps longer agent runs under control

Google Antigravity release notes have a new CLI signal after our latest Antigravity post: version 1.0.14, published on GitHub on June 30, 2026. The main app version on the Antigravity site still sits at 2.2.1, but the CLI track received changes worth noticing if you let an agent work for longer than a normal chat turn.

Google Antigravity is Google's agentic development environment: an IDE and CLI where a coding agent can read files, propose changes, run tools, and leave work for human review. An agentic IDE is not just an editor with chat. It is a workspace where the agent can do parts of the job with clear checkpoints. An approval gate is one of those checkpoints, where a human approves file changes, commands, or artifacts before the agent continues.

Google Antigravity release notes: CLI 1.0.14 in the changelog

Antigravity CLI 1.0.14 is a narrow but practical release. It allows image pasting from the clipboard in local tmux sessions, removes the max limit for the /goal command, lets subagents auto approve artifacts in an "always proceeds" mode when the parent agent is blocked, and fixes plugin imports, MCP configuration, TUI layout, and the inline view after rewind.

MCP, Model Context Protocol, is a way to connect an agent to external tools and data sources through a shared structure. That makes the small MCP fix more important than it first looks: if the CLI and permission manager do not read the same configuration path, a planned integration can fail to load or behave differently across environments.

Source: Antigravity CLI 1.0.14 on GitHub and Antigravity CLI CHANGELOG.md.

What actually changed

For practical teams, this is mainly an operations release for longer agent runs:

• Local tmux sessions can accept pasted images from the clipboard. That helps when an agent needs to review a screenshot, an error view, or visual source material without a separate file workflow.
• /goal no longer has a max limit according to the release notes. That fits work that should run until completed or cancelled, but it also raises the need for good intermediate receipts.
• Subagents can auto approve artifacts in "always proceeds" mode so they do not hang when the parent agent is blocked.
• Plugin import now copies the whole plugin directory, including folders outside skills/, such as shared/.
• A MCP configuration mismatch between the CLI and permission manager is fixed, which should make custom MCP servers more reliable.
• TUI race conditions and the inline viewport after rewind received bug fixes.

Source: GitHub release 1.0.14.

Why Swedish and Nordic teams should care

This is not a release that demands a new AI project. It works better as a control step inside a workflow you already test.

If you already use Antigravity CLI for code, documentation, debugging, or internal automation, 1.0.14 can help in two ways. First, source material gets broader when images in local tmux sessions are easier to bring in. Second, longer agent jobs become more realistic when /goal can continue until the job is completed or cancelled.

That also means you should write clearer work receipts. Ask the agent to report which sources it read, which files it changed, which artifacts subagents approved, and which commands still need a human. Integrate with real tools when useful, but use scoped permissions, secrets in env vars or a secret manager, redaction of sensitive values, and logs that someone can review afterward.

Short example: use the new Antigravity feature

Human step: open the right project in Google Antigravity CLI 1.0.14. If you use local tmux and have an image as source material, paste the image into the session yourself. Do not start a long /goal run until you know what the agent should return for review.

You are already running in the correct Google Antigravity CLI project.
Prepare a longer agent task that can use 1.0.14.
Use any pasted image as source material, check which MCP tools the project expects, and split the work into reviewable receipts.
Return: goal, sources checked, proposed steps, artifacts subagents can produce, and which commands or permissions need human approval.

Good output should:

• show which files, images, or MCP sources the agent actually checked,
• separate agent work from human approvals,
• break one long goal into small reviewable receipts,
• mention whether any plugin or MCP configuration should be checked before the run.

What to watch next

The public Antigravity app changelog still shows 2.2.1 as the latest app release, with gradual rollout wording on Google's releases page. That makes today's Antigravity post narrow: CLI 1.0.14, not a new app version. It is still a useful signal. Google is not only polishing the agent's answers, but the work pattern around the agent: image sources, longer goals, subagents, plugins, and MCP.

Source: Google Antigravity changelog, Google Antigravity releases, and Antigravity CLI 1.0.14.