Google Antigravity release notes: 2.6.0 and CLI 1.1.11 make long agent runs easier to inspect

Google Antigravity 2.6.0 and CLI 1.1.11 solve a fairly ordinary problem: agent runs become hard to operate when history grows, background jobs vanish from view, or status checks require manual detective work. Both releases are dated August 7. The app rolls out gradually.

Google Antigravity is Google's agentic development environment, where AI agents can work with project files, terminals, connected tools, and automations. A hook is a rule-driven function that runs at a defined point in an agent's lifecycle, such as after a tool call or when a turn ends.

Google Antigravity release notes: what changes in 2.6.0?

Antigravity 2.6.0 opens long conversations much faster and speeds up the conversation list for users with a large history. Conversations started by background automations now appear under the scheduled-tasks filter instead of disappearing from the sidebar. That makes automation work easier to find when a person needs to review the result.

Google also fixed several hook and subagent problems. Hook configurations that can never run are rejected when they load. Model-based hooks stop at their configured timeout with a clear error, and custom hooks now run at the end of a turn. Stopping a subagent also stops nested subagents and their background tasks.

Organizations get an important correction for connected tool servers: administrator policies now apply correctly at startup. MCP, the Model Context Protocol, is a standard way to connect an AI client to external tools and data sources.

Source: Google Antigravity changelog, version 2.6.0.

Antigravity CLI 1.1.11 reports status without an agent turn

CLI 1.1.11 can answer the read-only /usage, /quota, /credits, /model, /effort, and /skills commands in print mode. Print mode is the CLI's non-interactive path for scripts and automations. The commands emit one tab-separated record per line or structured data with --output-format json or stream-json. They do not start an agent turn, spend quota, or leave a conversation behind.

Use this check before a scheduled run to inspect the model, reasoning effort, available skills, and usage before the job continues. Interactive-only slash commands now fail explicitly in print mode instead of reaching the model as ordinary prompt text.

The release also adds a Vim mode under /settings and Editor Mode. It is off by default and includes Normal, Insert, Visual, and Visual Line modes, plus Vim editing in diff and artifact comments.

Source: Google Antigravity CLI 1.1.11 release notes.

Human step: capture a status snapshot

First update to Antigravity CLI 1.1.11 through your usual distribution process. Then run the read-only checks your workflow needs:

antigravity -p "/model" --output-format json
antigravity -p "/effort" --output-format json
antigravity -p "/skills" --output-format json
antigravity -p "/usage" --output-format json

The -p and --output-format flags, along with the json and stream-json values, are present in the distributed CLI 1.1.11 help. The slash commands above are the read-only print-mode commands listed by Google.

Source: Google Antigravity CLI 1.1.11 release notes.

Short example: use the new Antigravity feature

Paste the status records into an Antigravity conversation and use this short instruction:

Review the model, effort, skills, and usage status records. Return READY or BLOCKED, list mismatches with the planned workflow, recommend one correction, and name the human approval required. Do not change files or settings.

Good output should:

• Separate verified values from missing information.
• Identify concrete mismatches without changing the environment.
• Leave a clear decision and next human check.

A developer note on SDK 0.1.10

Google also documented Antigravity SDK 0.1.10 on August 4. It adds live UsageUpdate events for token use, call_id correlation across tool executions and hooks, and HookContext for sharing state across lifecycle callbacks. That matters for teams building their own monitoring, but app 2.6.0 and CLI 1.1.11 have broader value for today's Antigravity users.

Source: Google Antigravity changelog, SDK 0.1.10.

Nordic teams running recurring AI work can put the status check before a scheduled agent run and let a person own any decision about the model, permissions, or continuation. A Tool Forge engagement can build the same check into the workflow as a logged review point.