antigravity-skills-manager

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads content from the vendor's repository at github.com/rmyndharis/antigravity-skills.
  • [REMOTE_CODE_EXECUTION]: Executes the official vendor CLI tool via npx, involving remote code retrieval and execution.
  • [COMMAND_EXECUTION]: Runs a local management script (skills_cli.py) to handle file system operations.
  • [PERSISTENCE]: Installs instruction files into the agent's persistent configuration path (~/.gemini/antigravity/skills/), which is automatically loaded in new sessions.
  • [INDIRECT_PROMPT_INJECTION]: Acts as a surface for instructions from a remote source to enter the agent context.
    1. Ingestion points: Remote skill instruction files from the rmyndharis catalog.
    2. Boundary markers: The instructions explicitly require user confirmation and ID verification before installation.
    3. Capability inventory: Writing files to the skills directory and making network requests via urllib.
    4. Sanitization: Recommends human review of catalog search results before proceeding with installation.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 06:53 AM
Security Audit — agent-trust-hub — antigravity-skills-manager