dogfood

Pass

Audited by Gen Agent Trust Hub on Mar 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the agent-device CLI tool and standard Unix commands (mkdir, cp) to manage testing sessions, capture device state, and organize report artifacts.
  • [EXTERNAL_DOWNLOADS]: The skill configuration allows for the use of npx agent-device, which may download the automation tool from the npm registry if it is not already available in the environment.
  • [PROMPT_INJECTION]: An indirect prompt injection surface exists as the skill ingests untrusted UI data from third-party mobile applications through the snapshot command. This represents a standard risk for agents interacting with external interfaces.
  • Ingestion points: UI tree and text data are ingested via agent-device --session {SESSION} snapshot -i in SKILL.md.
  • Boundary markers: No specific delimiters are used for the UI snapshot data.
  • Capability inventory: The skill can execute shell commands, write files to the local system (screenshots, reports, videos), and interact with mobile devices via the agent-device tool.
  • Sanitization: The skill does not explicitly sanitize the text content retrieved from the mobile application's UI before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 2, 2026, 04:13 PM