The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from mobile application user interfaces, which represents untrusted external data that could contain malicious instructions.
Ingestion points: Data is ingested through tools like describe, screenshot, and debugger-component-tree used throughout the interaction loop in SKILL.md.
Boundary markers: The instructions do not define specific delimiters or instructions for the agent to disregard prompts embedded within the UI of the application under test.
Capability inventory: The skill provides tools for active manipulation such as keyboard, gesture-tap, and flow-execute, which could be exploited if an application displays malicious instructions.
Sanitization: There are no explicit requirements mentioned for sanitizing or filtering text content extracted from the app UI before the agent uses it for decision-making.
[SAFE]: The skill documentation explicitly mandates a secure credential management pattern, using {{secret:NAME}} placeholders for authentication to ensure that plaintext passwords never enter the agent's context or conversation logs.
[SAFE]: Tool usage and setup steps, including the use of adb and simulator management tools, are standard for the mobile development and testing domain and are consistent with vendor-provided infrastructure.

argent-test-ui-flow