Antigravity 1.2.10

Released: 2026-09-24

Release notes

FEATURE

/config Adds Medium Verbosity to Summarize Tool Calls While Keeping Commands and Responses Visible

When delegating complex tasks to an agent, you might want to avoid being flooded with pages of tool calls and thought details while still clearly tracking executed commands and final responses. The /config settings menu now introduces a medium verbosity option between high and low. In this mode, the agent groups a series of related tool calls and thought processes into concise summaries like Explored N files. All executed commands and resulting text responses remain fully visible on the screen. The Verbosity setting and each individual option beneath it now include descriptive text clearly outlining what is shown or hidden.

FEATURE

Artifact Viewer Accelerates Diagram Rendering and Falls Back to Text Diagrams in Multiplexers

Viewing diagrams or math formulas in the terminal is now smoother across loading and key interactions. On terminals supporting the Kitty graphics protocol, the artifact viewer pre-renders artifacts in the background so Mermaid diagrams and LaTeX formulas appear instantly upon opening. Images are uploaded once at a smaller size, eliminating flickering when zooming with Ctrl+= and Ctrl+-. The viewer footer combines the separate top and bottom hints into g/G top/bottom, labels the m key with the active switch target (including diagram, LaTeX, ASCII, or raw), and fixes hints containing arrow symbols wrapping too early. Inside tmux or GNU screen where images cannot reach the terminal, the system automatically falls back to ASCII text diagrams to prevent blank gaps, while setups with passthrough can still force image mode using CLI_GRAPHICS=kitty.

FEATURE

Terminal Sandbox Reduces Bypass Requests with Refined Failed Step Labels and Esc Key Handling

Running commands automatically with an agent in the terminal is now safer and more intuitive. With improvements to the terminal sandbox, the agent requests to bypass the sandbox significantly less often. The agent now recognizes that sandboxed commands can directly read and write its own artifact and scratch directories, and it continues to prioritize the sandbox for subsequent commands even after an earlier command required a bypass. For commands that finish execution with a non-zero exit code, the step title is formally updated from Errored to Failed. When the suggestions dropdown is open in the terminal, pressing Esc now closes the dropdown first, requiring a second Esc press to interrupt the agent and preventing accidental cancellation during autocomplete.

FEATURE

Customization Configs Now Scan Top-Level Directories Only, Requiring Nested Items via include_only

Loading behavior for customization files is now more precise and predictable when defining skills, rules, or plugins directories in a project. In customization files such as skills.json, rules.json, agents.json, and plugins.json, directory entries now load only first-level items located directly inside that directory—matching the behavior of the default .agents/skills/ directory—rather than recursively scanning every subdirectory. To include a specific nested item within a deeper subdirectory, you must explicitly specify its path via include_only, for example:

{"path": "shared_skills", "include_only": ["category/my-skill"]}

BUGFIX

Headless Mode Streaming Errors Now Return Exit Code 3 While JSON Output Retains Partial Responses

When running the CLI in automation scripts, silent failures that masquerade as success are a major pitfall. In headless mode executed with -p or --prompt, streaming outputs interrupted by model or agent errors previously exited with code 0 indicating success. Such failures now properly exit with code 3 and print an AGY_ERROR line, while JSON error output preserves the partial response generated prior to the failure, and multi-turn stream-json sessions continue to warn and proceed.

BUGFIX

Upgraded Legacy Remote Control Installations No Longer Crash Loop and Unregister Deprecated Services

Machines configured with legacy Remote Control installer scripts previously fell into infinite crash loops after upgrading. When the CLI is launched by that deprecated background service, it now automatically unregisters the background service and prompts switching to the remote-control start command, completely resolving crash loops after upgrades.

BUGFIX

Subagent Worktree Files Are No Longer Mistaken for Artifacts to Prevent Metadata Residue

When invoking background agents to work on code in parallel, project directories will no longer be cluttered with unexpected metadata files. Previously, writing standard files inside a Git worktree caused the system to mistake them for artifacts, demanding descriptions and leaving .metadata.json files behind. Subagent worktrees now reside under a unified worktrees/ folder in the app data directory, resolving this issue.

Original changelog

Original changelog/config Adds Medium Verbosity to Summarize Tool Calls While Keeping Commands and Responses Visible
Added `medium` verbosity mode to `/config`, between `high` and `low`, which groups related tool calls and thoughts into concise summaries (such as `Explored N files`) while keeping commands and responses visible; the Verbosity setting and each option now describe what they show
Original changelogArtifact Viewer Accelerates Diagram Rendering and Falls Back to Text Diagrams in Multiplexers
Improved Mermaid diagram and LaTeX rendering in the artifact viewer on Kitty-compatible terminals: artifacts are pre-rendered in the background so diagrams appear as soon as the viewer opens, images are uploaded once at a smaller size, and zooming with `Ctrl+=` / `Ctrl+-` no longer flickers between the old and new sizes
Improved the artifact viewer footer by combining the separate top and bottom hints into one `g/G top/bottom` entry, labeling `m` with the view it switches to (diagram, LaTeX, ASCII, or raw), and fixing hints that contain arrow symbols wrapping onto a new line too early
Fixed Mermaid diagrams and LaTeX leaving a blank gap inside `tmux` or GNU `screen` when the outer terminal supports Kitty graphics; the CLI now falls back to ASCII diagrams unless images actually reach the terminal, and `CLI_GRAPHICS=kitty` still forces image mode for setups with image passthrough configured
Original changelogTerminal Sandbox Reduces Bypass Requests with Refined Failed Step Labels and Esc Key Handling
Improved terminal sandbox behavior so the agent asks to bypass the sandbox less often: it now knows sandboxed commands can read and write its own artifact and scratch directories, and it tries the sandbox first even after an earlier command needed a bypass
Changed the step title of commands that exit with a non-zero code from `Errored` to `Failed`
Fixed pressing `Esc` while the suggestions dropdown is open interrupting the agent's turn; it now closes the dropdown, and a second `Esc` interrupts
Original changelogCustomization Configs Now Scan Top-Level Directories Only, Requiring Nested Items via include_only
Changed how directory entries in `skills.json`, `rules.json`, `agents.json`, and `plugins.json` are scanned: an entry now loads only the items directly inside the directory, the same as a `.agents/skills/` folder, instead of recursively loading everything beneath it; to load a nested item, name it in `include_only`, for example `{"path": "shared_skills", "include_only": ["category/my-skill"]}`
Original changelogHeadless Mode Streaming Errors Now Return Exit Code 3 While JSON Output Retains Partial Responses
Fixed headless (`-p` / `--prompt`) runs that streamed part of a response and then ended on a model or agent error exiting with code 0; they now exit with code `3` and print the `AGY_ERROR` line, and JSON error output includes the partial response, while multi-turn `stream-json` sessions still warn and continue
Original changelogUpgraded Legacy Remote Control Installations No Longer Crash Loop and Unregister Deprecated Services
Fixed machines set up with the old Remote Control installer script crash-looping after an upgrade; when the CLI is launched by that deprecated background service it now unregisters the service and points to `remote-control start` instead of restarting repeatedly
Original changelogSubagent Worktree Files Are No Longer Mistaken for Artifacts to Prevent Metadata Residue
Fixed files that subagents write inside their own Git worktree being treated as artifacts, which demanded artifact metadata and left `.metadata.json` files in the worktree; subagent worktrees now live under a `worktrees/` directory in the app data directory