mobile-harness
Pass
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill is configured to maintain its own environment by performing self-updates through the vendor's repository and installing Python package updates.
- Evidence: AGENTS.md and UPDATE.md include instructions for the agent to run
git pull,npx skills update, andpip install -Uon vendor-owned packages likemobilerun-core. - [INDIRECT_PROMPT_INJECTION]: The skill interacts with external mobile device screens and web content, which are treated as untrusted data inputs.
- Ingestion points: Application UI data is ingested via
device.ui(),device.find_nodes(), and browser-based JavaScript execution results. - Boundary markers: The skill includes explicit non-negotiable rules in AGENTS.md and core/memory/GUIDE.md that screen content must never be treated as agent instructions.
- Capability inventory: The skill has access to local file storage, shell command execution for device management, and network communication with vendor cloud endpoints.
- Sanitization: Guarded by explicit instructions to the agent to maintain the distinction between device data and operational logic.
Audit Metadata