The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to call its internal AI processing logic. The command is passed as a list of individual strings, which ensures that user input cannot be interpreted as shell commands, effectively preventing command injection.
[EXTERNAL_DOWNLOADS]: The script communicates with OpenRouter and Perplexity AI via their official API endpoints. These are well-known services used as intended for research and image generation, and no unauthorized or suspicious third-party downloads were detected.
[PROMPT_INJECTION]: The skill ingests user input and research data which are interpolated into larger system prompts.
Ingestion points: The prompt command-line argument and the results from the Perplexity 'sonar-pro' research model.
Boundary markers: The code uses structural headers such as 'USER REQUEST:' to separate different data types within the prompt.
Capability inventory: The skill is limited to writing image files and JSON logs to the local directory and making authenticated API calls.
Sanitization: The implementation relies on the safety filters of the integrated LLM providers to handle potential malicious content in the input text.

infographics