The Agent Skills Directory

[COMMAND_EXECUTION]: The script generate_slide_image.py uses subprocess.run to orchestrate the internal AI generation logic contained in generate_slide_image_ai.py. This is a standard modular script execution pattern within a skill's own directory.
[COMMAND_EXECUTION]: The validate_presentation.py script invokes pdflatex via subprocess.run to verify the compilation of LaTeX Beamer source files, which is a core feature for academic presentation validation.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to openrouter.ai to access Large Language Models for generating slide images and performing quality reviews. This interaction is central to the skill's AI-assisted slide generation workflow.
[INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided slide descriptions as input for the AI generation prompt. 1. Ingestion points: The prompt parameter in generate_slide_image.py and generate_slide_image_ai.py acts as the entry point for untrusted data. 2. Boundary markers: The script uses structured templates like FULL_SLIDE_GUIDELINES to encapsulate user input, providing context to the LLM on how to interpret the request. 3. Capability inventory: The skill has file system access and shell execution capabilities used to manage presentation assets and run validation tools. 4. Sanitization: While there is no explicit character escaping, the input is processed by a secondary review LLM to ensure the output meets quality standards.
[CREDENTIALS_UNSAFE]: The documentation and scripts correctly instruct users to manage the OPENROUTER_API_KEY via environment variables or .env files, following best practices for secret management and avoiding hardcoded credentials.

scientific-slides