Antigravity Deep Research

Run $ARGUMENTS as a deep-research query on Gemini's deep-research managed agent, wait for it to finish, and report the cited result. Two models are available: deep-research-preview-04-2026 (default) and deep-research-max-preview-04-2026 (max mode — slower, higher effort), selected with --max.

The script is at ${CLAUDE_PLUGIN_ROOT}/scripts/antigravity.py. It is self-daemonizing: research returns immediately with a run_id, a detached worker runs the research, and a status file flips to completed / failed when done. Requires credentials and uv on PATH. Two credential paths are supported: Vertex AI via ADC plus ANTIGRAVITY_PROJECT / ANTIGRAVITY_LOCATION, or an API key (GEMINI_API_KEY / GOOGLE_API_KEY); see the Authentication section of ../delegate/references/usage.md. Deep research can take several minutes.

Phase 1: Frame the query

Goal: Turn $ARGUMENTS into a clear research question.

Actions:

1. Use the full $ARGUMENTS text (minus a --max flag) as the research question. Pass --max through to the script when the user wants max mode (deeper, slower).
2. If it is empty or too vague to research (no subject, scope, or constraints), ask the user one or two clarifying questions, then proceed.

Phase 2: Launch the run

Goal: Start the detached research worker.

Actions:

1. Run the script:

   uv run "${CLAUDE_PLUGIN_ROOT}/scripts/antigravity.py" research --query "<question>" [--max]
   

2. Capture run_id, output_file, and wait_command from stdout.
3. If stdout reports an error (for example no credentials configured, or a forced ANTIGRAVITY_AUTH path that is incomplete), surface it and stop. Note: deep research has no token cap (DeepResearchAgentConfig accepts no max_total_tokens), so --max-tokens is deliberately absent here — never imply to the user that a research run is bounded.

Phase 3: Wait for completion

Goal: Block until the research finishes without busy-looping the model.

Actions:

1. Start a Monitor on the captured wait_command. It emits one line — antigravity run <id>: completed / failed / timeout — then exits:

   uv run "${CLAUDE_PLUGIN_ROOT}/scripts/antigravity.py" wait --run <run_id> --timeout 1800
   

   Deep research is slow: set the Monitor timeout_ms to 3600000 (60 min, 2x the wait timeout) and pass --timeout 1800 to the wait command. Use a clear description like "antigravity research <run_id>".
2. When the Monitor event arrives, check if the line contains : completed, : failed, or : timeout:
   • Contains : completed or : failed → proceed to Phase 4.
   • Contains : timeout → the research is still running (the worker keeps polling up to ~2h). Start the Monitor on the same wait_command again to keep waiting. After four consecutive timeouts (4 hours total), tell the user it is still running and give them the full command to fetch it later:

     uv run "${CLAUDE_PLUGIN_ROOT}/scripts/antigravity.py" status --run <run_id> --full
     

     then stop. Never present a timeout / still-running state as the report.

Phase 4: Report the result

Goal: Present the research report.

Actions:

1. Fetch the full result:

   uv run "${CLAUDE_PLUGIN_ROOT}/scripts/antigravity.py" status --run <run_id> --full
   

   Or read the rendered output_file directly.
2. Present the report text to the user, preserving any citations and sources.
3. If the status is failed, report the recorded error and likely cause.

Notes

• This uses Gemini's managed deep-research agent; it manages its own search and browsing. No sandbox tools or network flags apply here.
• See ../delegate/references/usage.md for the shared API surface and limits.