The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands using the subprocess module to handle PPTX rendering (via LibreOffice soffice, macOS Keynote via osascript, or Windows PowerPoint COM) and for the optional codex image generation backend. These operations are intended for the skill's primary functionality and use established system utilities.
[INDIRECT_PROMPT_INJECTION]: The skill accepts slide outlines and content from users, which it interpolates into prompts for external AI models. This creates an indirect prompt injection surface; however, the risk is mitigated as the output is restricted to image generation and static PPTX files, minimizing potential impact on the host system.
[EXTERNAL_DOWNLOADS]: The image_generator.py script downloads images from URLs provided by the AI model's API. The implementation includes safety checks, such as verifying the protocol is strictly http/https and enforcing a 50MB download limit to prevent potential denial-of-service or malicious file delivery.
[CREDENTIALS_SAFE]: The skill implements a secure credential management strategy. It explicitly avoids recursive searches for .env files in parent directories, instead relying on platform-injected variables or specific skill-owned configuration paths, which prevents the accidental exposure of sensitive project-level secrets.

gpt-image2-ppt