The Agent Skills Directory

[SAFE]: The skill functions as a visualization orchestrator, reading architectural data from the 'AI
Project Brain' directory and generating renderings stored in '10
Project Photos'. All operations, including calls to the 'image-generation-mcp' server, align with the stated primary purpose.
[PROMPT_INJECTION]: The system architecture exhibits a surface for indirect prompt injection (Category 8) by interpolating untrusted project data into generated prompts. This is documented as a structural risk rather than an active exploit. \n
Ingestion points: Reads project-specific metadata and site context from multiple JSON files in the 'AI
Project Brain/' folder (SKILL.md). \n
Boundary markers: Templates in the 'references/' directory utilize standard placeholder substitution ({{VARIABLE}}) but lack explicit delimiter markers or 'ignore' instructions for interpolated content. \n
Capability inventory: The skill can execute image generation through external APIs (Gemini, Flux 2) and perform local file writes to the rendering gallery and logs. \n
Sanitization: The documentation describes a variable substitution engine (prompt-assembler.js) but does not specify any sanitization or validation routines for the incoming project data.

rendering-generator