The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface where malicious instructions could be embedded in the data the agent processes.
Ingestion points: The skill instructs the agent to fetch and analyze 20–30 real mobile app screenshots and videos via the 'Appllama MCP' to study UI patterns (SKILL.md).
Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat this external visual/textual content as untrusted or to disregard instructions potentially found within those screenshots.
Capability inventory: The agent has the capability to generate and modify React Native code, write to the filesystem, and execute shell commands for simulator interaction (xcrun simctl).
Sanitization: No sanitization, OCR filtering, or validation of the content extracted from the screenshots is mentioned or enforced.
[COMMAND_EXECUTION]: The skill provides instructions for executing shell commands to interact with the iOS simulator environment.
Evidence: The file references/simulator-loop.md directs the agent to use xcrun simctl io booted screenshot and xcrun simctl io booted recordVideo for UI and motion verification.
[EXTERNAL_DOWNLOADS]: The skill references external Model Context Protocol (MCP) tools and CLI utilities for its core functionality.
Evidence: SKILL.md and references/image-assets.md mention the 'Appllama MCP' and 'Higgsfield MCP/CLI' for research and asset generation. These tools originate from the skill's associated vendor environment.

appllama-app-design-skill