skills/abpai/skills/antigravity/Gen Agent Trust Hub

antigravity

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/antigravity-run.sh executes the agy (Antigravity) CLI binary. It allows users to pass extra arguments directly to the binary via the -- separator and provides an --allow-all flag that maps to the CLI's internal permission bypass.
  • [DYNAMIC_EXECUTION]: The runner script generates runtime helper scripts (monitor.sh and continue.sh) and executes embedded Python snippets to manage JSON processing, environment variable parsing, and process lifecycle controls. These generated scripts use shell-quoted variables to prevent injection.
  • [INDIRECT_PROMPT_INJECTION]: The skill acts as an interface for the Gemini-powered Antigravity CLI, creating a surface for indirect injection via processed repository content or external prompts.
  • Ingestion points: Prompt text/files and local workspace content provided via command-line arguments in scripts/antigravity-run.sh.
  • Boundary markers: The skill implements a 'review' mode that uses an isolated Git clone of the workspace to prevent accidental writes to the source repository.
  • Capability inventory: The CLI worker can perform file reads/writes, execute shell commands, and make network calls to Google services (controlled by the --dangerously-skip-permissions and --sandbox flags).
  • Sanitization: The runner fingerprints the source workspace before and after execution to detect and fail on unauthorized modifications.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 03:17 PM