Antigravity helper

antigravity-helper

This skill lets any Anthropic-Skill-aware agent (Codex CLI, OpenCode, Pi.dev, plus Claude Code via .claude/skills/) delegate to the Antigravity CLI (agy) for code review and task analysis. It wraps the antigravity-cli-plugin-cc companion script — a stateless Node.js dispatcher that assembles the prompt, writes it to a temp workspace directory, applies read-only or write-capable execution, runs agy in print mode, and extracts the marker-wrapped response so agy's tool-use narration never reaches the user.

If you are running inside Claude Code, prefer the slash-command surface (/ask-antigravity:review, /ask-antigravity:adversarial-review, /ask-antigravity:rescue, /ask-antigravity:setup) provided by the antigravity-cli-plugin-cc plugin. This skill is the portable fallback for agents without that plugin format.

Prerequisites

• agy CLI 1.0.7 or later, installed and authenticated. Install with curl -fsSL https://antigravity.google/cli/install.sh | bash or brew install --cask antigravity-cli (agy is not distributed via npm). Older agy hangs in headless print mode; the companion refuses to invoke it and asks for an upgrade. Sign in by running agy interactively and completing Google sign-in, or set ANTIGRAVITY_API_KEY in the environment.
• node 18.18 or later on PATH.
• ANTIGRAVITY_CLI_PLUGIN_CC_ROOT env var pointing at the absolute path of the cloned antigravity-cli-plugin-cc repository. If that variable is unset, ask the user for the path before running the companion.
• Workspace trust. If agy has not been interactively trusted in the directory you invoke it from, trust the workspace in agy (it records trustedWorkspaces in its settings.json). Without trust, headless agy may refuse to proceed.

When to invoke this skill

Trigger on requests that match any of:

• "review my changes / diff / branch with Antigravity"
• "challenge this design choice / pressure-test this implementation"
• "ask Antigravity to investigate / analyze / fix X"
• "give me a second opinion on…" (especially when the question benefits from a large-context single pass)

Do not trigger when the user only wants Claude's own review or when the task is small enough that delegation adds latency without value.

Invocations

All commands assume the working directory is the repo being reviewed/analyzed.

Code review (read-only, structured markdown output)

node "$ANTIGRAVITY_CLI_PLUGIN_CC_ROOT/plugins/ask-antigravity/scripts/antigravity-companion.mjs" review

node "$ANTIGRAVITY_CLI_PLUGIN_CC_ROOT/plugins/ask-antigravity/scripts/antigravity-companion.mjs" review --base main

The output is markdown with a fixed skeleton: ## Summary then ### Critical / High / Medium / Nits sections. Stream it back to the user verbatim.

Adversarial review (read-only, steerable, accepts focus text)

node "$ANTIGRAVITY_CLI_PLUGIN_CC_ROOT/plugins/ask-antigravity/scripts/antigravity-companion.mjs" adversarial-review "look for race conditions in the retry path"

Same output shape; framing is "find reasons this should not ship" rather than general review.

Delegate a task / second opinion

# Default: Antigravity analyzes and proposes. NOTE: this is not a sandbox — agy may
# still edit files (see "Not a guaranteed sandbox" below). Run on a clean git tree.
node "$ANTIGRAVITY_CLI_PLUGIN_CC_ROOT/plugins/ask-antigravity/scripts/antigravity-companion.mjs" task "investigate why the build is failing in CI"

# Write intent: passes --dangerously-skip-permissions so agy won't stall on prompts.
node "$ANTIGRAVITY_CLI_PLUGIN_CC_ROOT/plugins/ask-antigravity/scripts/antigravity-companion.mjs" task --write "fix the failing test with the smallest safe patch"

Prefer --prompt-file <path> over inline text when the task text is untrusted or contains shell metacharacters: it carries the text as data instead of on a shell command line.

Setup probe (verify install + auth)

node "$ANTIGRAVITY_CLI_PLUGIN_CC_ROOT/plugins/ask-antigravity/scripts/antigravity-companion.mjs" setup
node "$ANTIGRAVITY_CLI_PLUGIN_CC_ROOT/plugins/ask-antigravity/scripts/antigravity-companion.mjs" setup --json

Note for Codex CLI users with Antigravity OAuth (not ANTIGRAVITY_API_KEY): the live review call may stall on an interactive browser auth prompt even though the setup probe reports authenticated: true. Two sandbox restrictions cause this — Codex's workspace-write mode (1) blocks outbound network, so agy's OAuth-token refresh HTTP request fails, and (2) blocks writes outside the workspace, so agy cannot persist the rotated token to its config under ~/.gemini/antigravity-cli/. Either failure makes agy fall back to interactive browser auth, which Codex's process isolation cannot complete. Both must be granted:

codex exec -s workspace-write \
  -c 'sandbox_workspace_write.writable_roots=["/Users/<you>/.gemini"]' \
  -c 'sandbox_workspace_write.network_access=true' \
  --skip-git-repo-check \
  "your prompt that triggers antigravity-helper"

The misleading part: the error reads like a user cancellation. The user did not cancel; the sandbox blocked the refresh. (Note: ~/.gemini/antigravity-cli/ is agy's real config directory — that path is not a typo.)

If the result reports installed: false, suggest installing agy with curl -fsSL https://antigravity.google/cli/install.sh | bash or brew install --cask antigravity-cli. If authenticated: false, suggest running agy interactively to sign in.

Model selection

Reviews and rescue runs default to the model currently selected in the agy TUI (/model picker). Pass --model "<display name>" to any subcommand to override per call — valid names come from agy models (e.g. Gemini 3.5 Flash (Low)); they contain spaces, so quote them.

Backgrounding

For large reviews (multi-file diffs, branch reviews, or long rescue tasks), run the companion in the host agent's background-bash facility (run_in_background: true for Claude Code, equivalent for other agents). The companion script itself is stateless and synchronous — it does not manage background jobs.

What this skill is NOT

• Not a slash-command surface. Skills are model-invoked; for /ask-antigravity:review-style UX, install the antigravity-cli-plugin-cc plugin into Claude Code instead.
• Not stateful. No transcripts, no session resume, no PID tracking. Every invocation is a fresh one-shot agy call.
• Not a guaranteed sandbox. review and adversarial-review are safe by construction: the whole diff is inlined into the prompt and agy is run in an isolated empty working directory, so it never sees the repo. But task runs agy in the repo, and omitting --write does not stop it from editing files — on agy 1.1.1 no CLI flag does (verified: omitting --dangerously-skip-permissions, --mode plan, and --sandbox all still wrote). --write signals intent and auto-approves prompts; it is not a capability boundary. Run task on a clean git tree so any change is visible in git diff.

Reference

• Source: antigravity-cli-plugin-cc repository, plugins/ask-antigravity/scripts/antigravity-companion.mjs
• License: Apache-2.0
• Test suite: node --test tests/*.test.mjs from the repo root (covering arg parsing, prompt assembly, git context collection, prompt-injection sanitization, symlink-exfiltration prevention, output capture/timeout handling, process-signal handling, and an end-to-end companion run against a fake agy). AGY_LIVE=1 node --test tests/live.test.mjs smoke-tests the real agy after upgrades.