The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess module to orchestrate local Python scripts and execute the pdflatex compiler. These calls are performed securely by passing arguments as lists rather than shell strings, which prevents shell injection. These operations are essential for the skill's functionality in validating and compiling scientific slides.
[INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided text which is interpolated into prompts for an external LLM. This creates a surface for indirect prompt injection; however, the risk is mitigated because the primary output is a static image intended for a presentation, which does not feed back into the agent's execution logic or system commands. The risk level is consistent with standard AI-based content generation tools.
[EXTERNAL_DOWNLOADS]: The implementation relies on well-known and reputable Python packages such as requests, Pillow, PyMuPDF (fitz), and python-pptx. These are standard libraries for HTTP requests, image processing, and document manipulation. The skill also connects to OpenRouter, a recognized service for LLM API access, to perform its generative tasks.
[CREDENTIALS_UNSAFE]: The skill requires an API key for its generation features but correctly manages it via environment variables or .env files. This aligns with secure development practices and avoids the risks associated with hardcoded credentials.

scientific-slides