Skip to content
Proposals/Add include_only for nested Antigravity skills,
proposalteamP5Worth a lookGoogle Antigravity

Add include_only for nested Antigravity skills, rules, and agents

Antigravity 1.2.10 loads only direct children of directory entries in skills.json, rules.json, agents.json, and plugins.json—name nested items in include_only. Set /config verbosity to medium, fail CI on headless exit code 3 (AGY_ERROR), and replace the deprecated Remote Control installer service with remote-control start.

Why this loop

1.2.10 no longer recursively loads directory entries in skills.json, rules.json, agents.json, and plugins.json. Each path now loads only items directly inside that directory, matching .agents/skills/. Nested items that used to appear automatically will vanish unless named in include_only, e.g. {"path": "shared_skills", "include_only": ["category/my-skill"]}. Update every folder entry in those four files the same way.

Set /config Verbosity to medium (between high and low). Related tool calls and thoughts collapse into summaries such as Explored N files while commands and responses stay visible; the setting now describes what each option shows.

Headless -p/--prompt runs that stream then hit a model or agent error used to exit 0. They now exit 3, print AGY_ERROR, and put the partial response in JSON error output. Multi-turn stream-json still warns and continues. Point CI at exit 3.

Keep the sandbox enabled: sandboxed commands may read/write the agent's artifact and scratch dirs, and the agent retries the sandbox first after a prior bypass. Subagent Git worktrees moved under worktrees/ in app data and are no longer treated as artifacts. Stop the deprecated Remote Control installer service and use remote-control start so upgrades do not crash-loop.

Proposed actions

  1. Update skills.json, rules.json, agents.json, and plugins.json: for each directory path that used recursive nested loading, add include_only listing those nested items, e.g. {"path": "shared_skills", "include_only": ["category/my-skill"]}. Only items directly inside the path load now.
  2. In the Antigravity CLI, run /config and set Verbosity to medium so related tool calls and thoughts group into concise summaries such as Explored N files while commands and responses remain visible.
  3. Change CI and scripts that run Antigravity with -p or --prompt to treat exit code 3 as failure, parse the AGY_ERROR line, and read the partial response from JSON error output; a streamed response that then errors is no longer exit 0.
  4. If this host still launches the CLI from the old Remote Control installer background service, stop that service and start with remote-control start so 1.2.10 unregisters it once instead of crash-looping.
  5. Leave the terminal sandbox on. 1.2.10 lets sandboxed commands read and write the agent's own artifact and scratch directories and tries the sandbox first even after an earlier bypass, so bypass prompts should be rarer.

Agent prompt

Paste into your agent or query via MCP (get_agent_prompt) — free, no extra AI cost

Google Antigravity / agent task

DevAgentRadar → Google Antigravity

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

Context

Assistant: Google Antigravity Proposal: Add include_only for nested Antigravity skills, rules, and agents Summary: Antigravity 1.2.10 loads only direct children of directory entries in skills.json, rules.json, agents.json, and plugins.json—name nested items in include_only. Set /config verbosity to medium, fail CI on headless exit code 3 (AGY_ERROR), and replace the deprecated Remote Control installer service with remote-control start. Primary source: https://github.com/google-antigravity/antigravity-cli/releases/tag/1.2.10

Why it matters

1.2.10 no longer recursively loads directory entries in skills.json, rules.json, agents.json, and plugins.json. Each path now loads only items directly inside that directory, matching .agents/skills/. Nested items that used to appear automatically will vanish unless named in include_only, e.g. {"path": "shared_skills", "include_only": ["category/my-skill"]}. Update every folder entry in those four files the same way.

Set /config Verbosity to medium (between high and low). Related tool calls and thoughts collapse into summaries such as Explored N files while commands and responses stay visible; the setting now describes what each option shows.

Headless -p/--prompt runs that stream then hit a model or agent error used to exit 0. They now exit 3, print AGY_ERROR, and put the partial response in JSON error output. Multi-turn stream-json still warns and continues. Point CI at exit 3.

Keep the sandbox enabled: sandboxed commands may read/write the agent's artifact and scratch dirs, and the agent retries the sandbox first after a prior bypass. Subagent Git worktrees moved under worktrees/ in app data and are no longer treated as artifacts. Stop the deprecated Remote Control installer service and use remote-control start so upgrades do not crash-loop.

Suggested actions

  1. Update skills.json, rules.json, agents.json, and plugins.json: for each directory path that used recursive nested loading, add include_only listing those nested items, e.g. {"path": "shared_skills", "include_only": ["category/my-skill"]}. Only items directly inside the path load now.
  2. In the Antigravity CLI, run /config and set Verbosity to medium so related tool calls and thoughts group into concise summaries such as Explored N files while commands and responses remain visible.
  3. Change CI and scripts that run Antigravity with -p or --prompt to treat exit code 3 as failure, parse the AGY_ERROR line, and read the partial response from JSON error output; a streamed response that then errors is no longer exit 0.
  4. If this host still launches the CLI from the old Remote Control installer background service, stop that service and start with remote-control start so 1.2.10 unregisters it once instead of crash-looping.
  5. Leave the terminal sandbox on. 1.2.10 lets sandboxed commands read and write the agent's own artifact and scratch directories and tries the sandbox first even after an earlier bypass, so bypass prompts should be rarer.

Config surfaces this release may change

  • skills (high confidence) — check your repo before applying
  • background and headless runs — 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-1-2-10-add-include-only-for-nested-antigravity-skills and mark Applied, Skipped, or Failed. Proposal id: 14cc6574-bb09-42c2-a20a-8cb8370bb838

Please:

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

Your loop

This browser · no sign-in · not shared as “you”

After you run the prompt

Only you can mark this. Agents cannot write your loop.

Your decision stays on this device. A public tally appears after a few votes.

Originating release signal

Google Antigravity1.2.10Sep 24, 2026

1.2.10

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 · +10 more changes
Verified excerpt — the source's own words
  • 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
  • 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
  • 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

Excerpt ends here — this release continues at the source ↗.

Primary source ↗