Skills
Navigate
skills/svpl-dev/my-agent-skills/agent-platform-migrate-from-ai-studio/Gen Agent Trust Hub

agent-platform-migrate-from-ai-studio

Fail

Audited by Gen Agent Trust Hub on Jun 13, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: Fetches and executes a configuration script from Google's official cloud-samples-data repository to set up Application Default Credentials (ADC).
  • Evidence: bash <(curl -sSL https://storage.googleapis.com/cloud-samples-data/adc/setup_adc.sh) in SKILL.md.
  • [COMMAND_EXECUTION]: Provides multiple shell commands for project environment setup, enabling Google Cloud APIs, and managing service account permissions using the gcloud CLI tool.
  • Evidence: Use of gcloud services enable, gcloud auth login, and gcloud projects add-iam-policy-binding throughout SKILL.md.
  • [PROMPT_INJECTION]: The skill configures and provides examples for agents (using OpenClaw and the GenAI SDK) that ingest untrusted user input directly into LLM prompts, representing a surface for indirect prompt injection.
  • Ingestion points: SKILL.md (demonstrating openclaw agent --message and SDK generate_content calls using user-controlled strings).
  • Boundary markers: Not present in the configuration examples to delimit instructions from user input.
  • Capability inventory: SKILL.md (configures tools like OpenClaw which can perform various automated tasks based on model output).
  • Sanitization: No input validation or output sanitization logic is provided in the configuration guide.
Recommendations
  • HIGH: Downloads and executes remote code from: https://storage.googleapis.com/cloud-samples-data/adc/setup_adc.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 13, 2026, 06:03 AM
Security Audit — agent-trust-hub — agent-platform-migrate-from-ai-studio