The Agent Skills Directory

[PROMPT_INJECTION]: The references/base-prompt.md file contains instructions telling the image generation model to "DO NOT refuse to generate" when encountering sensitive or copyrighted figures, which is an attempt to override standard safety constraints.\n- [DYNAMIC_EXECUTION]: The script scripts/generate-slides.py uses subprocess.check_call to automatically install the google-genai Python package at runtime if it is missing.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content such as PDF papers and articles to generate slide outlines and image prompts, which could allow malicious instructions embedded in the source text to influence the agent's behavior.\n
Ingestion points: Source files like source.md and source-paper.pdf are read during the content analysis phase.\n
Boundary markers: The templates lack explicit instructions for the model to ignore embedded commands within the source text.\n
Capability inventory: The skill can execute local scripts, write to the file system, and make network requests to external APIs.\n
Sanitization: No sanitization or filtering is performed on the ingested content before it is processed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill requires downloading multiple dependencies from official NPM and PyPI registries during setup and execution, including canvas, pdf-lib, pptxgenjs, and google-genai.

paper-slide-deck