The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in a wrapper script to manage the execution of its core AI generation logic. The command is constructed using a list of arguments rather than a raw string, which effectively mitigates shell injection risks.
[EXTERNAL_DOWNLOADS]: The skill communicates with openrouter.ai to access AI services provided by Google and Perplexity. This is a standard and well-known service for managing multiple AI models, and the network operations are limited to this specific API endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill processes user descriptions and external research data which are then used to build prompts for AI generation and review. This ingestion of untrusted external content represents a surface for indirect prompt injection.
Ingestion points: User-provided prompts and data gathered from Perplexity Sonar research.
Boundary markers: The current implementation does not use explicit delimiters to separate external data from system instructions within the generated prompts.
Capability inventory: The skill is permitted to write files (images and logs) and execute shell commands within its restricted environment.
Sanitization: No specific sanitization logic was found for the external text inputs before they are interpolated into the prompt templates.

infographics