skills/simota/agent-skills/voyager/Gen Agent Trust Hub

voyager

Pass

Audited by Gen Agent Trust Hub on Oct 22, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides extensive documentation and instructional content for authoring E2E test suites. No patterns of prompt injection, data exfiltration, or malicious persistence were identified.
  • [SAFE]: The skill relies on industry-standard, well-known testing frameworks and services (e.g., Playwright, Cypress, Appium, BrowserStack, Sauce Labs, AWS Device Farm). These are established technology providers, and the instructions for their use follow standard, safe patterns.
  • [SAFE]: The skill explicitly warns against hard-coding credentials and instead provides templates for secure secret management using environment variables and .env files, which is a security best practice.
  • [SAFE]: The use of administrative commands (such as sudo xcode-select) is limited to specific environment configuration tasks within CI/CD pipelines (e.g., GitHub Actions), which is standard operational procedure and not indicative of malicious privilege escalation.
  • [INDIRECT_PROMPT_INJECTION]: The skill inherently possesses an attack surface for indirect prompt injection as its primary function involves the agent reading and processing external data from web pages and accessibility trees.
  • Ingestion points: The agent ingests web page content, DOM structures, and API responses during test execution (e.g., ai-powered-e2e-testing.md, api-e2e-testing.md).
  • Boundary markers: The skill mitigates risks by implementing a mandatory 'AI Test Review Checklist' and human review gates before merging AI-generated tests (ai-powered-e2e-testing.md).
  • Capability inventory: The skill utilizes shell execution via npm/npx for test runners, file writing for test scripts, and network access for communication with tested endpoints and cloud device farms.
  • Sanitization: The skill prioritizes semantic targeting (getByRole) and accessibility-first testing methods which rely on stable accessibility trees rather than raw, potentially injectable DOM content.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 22, 2026, 01:03 PM