The Agent Skills Directory

[SAFE]: The skill manages sensitive API credentials by loading them from local environment files (e.g., .env in the PAI directory) rather than hardcoding them within the instruction files or scripts.
[SAFE]: External network communications are exclusively directed to trusted and well-known AI and image processing service providers, such as Google, OpenAI, Replicate, and remove.bg.
[COMMAND_EXECUTION]: The skill utilizes local shell commands (via magick, cwebp, and curl) to perform image transformations and provide status notifications. These executions are constrained to legitimate local utilities and do not present an external exfiltration or remote code execution risk.
[PROMPT_INJECTION]: The skill includes workflows that ingest external text content (e.g., essays) for content-aware visualization. While this creates a theoretical surface for indirect prompt injection, it is a functional requirement of the skill and is mitigated by the safety filters of the integrated LLM backends.

Art