antigravity-for-claude-code

Pass

Audited by Gen Agent Trust Hub on Jul 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes local shell and Python scripts, such as scripts/agy-delegate.sh, scripts/agy-job.sh, and scripts/measure-session.py, to orchestrate work between the agent and the CLI.
  • [COMMAND_EXECUTION]: It executes the agy command-line utility, which is a tool provided by Google for interacting with Gemini models.
  • [COMMAND_EXECUTION]: The skill documentation supports the use of a --yolo flag, which grants the model autonomy to execute terminal commands and use system tools.
  • [DATA_EXFILTRATION]: The skill reads workspace directories and individual files to provide context to the external model (Gemini) during task delegation.
  • [DATA_EXFILTRATION]: It includes functionality to retrieve and analyze production logs from Cloud Run services for debugging purposes.
  • [PROMPT_INJECTION]: The skill manages an attack surface for indirect prompt injection by ingesting external data while possessing high-privilege tool capabilities.
  • Ingestion points: Local source code files, git diffs provided via stdin, and remote Cloud Run service logs.
  • Boundary markers: No specific delimiters or "ignore instructions" markers are defined for the ingested data within the skill documentation.
  • Capability inventory: The skill can perform filesystem modifications (--mode accept-edits), execute arbitrary terminal commands (--yolo), and access network resources via the CLI.
  • Sanitization: No evidence of input validation or sanitization of processed content is provided.
  • [EXTERNAL_DOWNLOADS]: The skill references and fetches the Antigravity CLI from Google's official infrastructure.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 17, 2026, 12:32 AM
Security Audit — agent-trust-hub — antigravity-for-claude-code