The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: Phase 1 and Phase 3 of the skill involve extracting and analyzing user-provided app metadata and screenshots. This data is then used to generate design specifications, copy, and HTML code. The instructions do not specify boundary markers or sanitization requirements for this input data to prevent potentially embedded malicious instructions from influencing the agent's behavior or the generated output.
Ingestion points: App descriptions, features, and visual reference analysis from the user are processed in the initial phases (Phase 1, Phase 3).
Boundary markers: There are no specific delimiters or warnings used when interpolating external content into the agent's creative context.
Capability inventory: The skill possesses the capability to write files (HTML/CSS) and execute shell commands via npx.
Sanitization: No instructions are provided for escaping or validating user input before its use in generation or command assembly.
[COMMAND_EXECUTION]: The skill directs the agent to run shell commands such as npx playwright screenshot and npx remotion still in Phase 4. These commands use arguments derived from the project context, which could potentially be manipulated if user-provided strings are not properly sanitized by the execution platform.
[DYNAMIC_EXECUTION]: As its primary function, the skill generates HTML/CSS scripts and then executes them via a headless browser environment (Playwright) or a React composition engine (Remotion). Generating and executing dynamic content is a powerful capability that represents an inherent security surface.

appshot-images