The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill uses ELEVENLABS_API_KEY and GEMINI_API_KEY to authenticate with AI services. Instructions in references/pronounce-button-guide.md and references/tts-installation.md correctly advise users to store these in environment variables and explicitly warn against hardcoding or committing secrets.
[EXTERNAL_DOWNLOADS]: The skill fetches assets and data from trusted and well-known sources, including Wikimedia Commons, US Federal Government archives (USDA, EPA, USGS, etc.), ElevenLabs, and Google. These operations are essential for the skill's primary purpose of sourcing educational media and are documented neutrally.
[REMOTE_CODE_EXECUTION]: The guides recommend the use of established tools such as npx to execute the Marp CLI and standard package managers (pip, npm) to install well-known libraries like google-genai, elevenlabs, and pptxgenjs. This constitutes standard development practice for integrating verified third-party tools.
[COMMAND_EXECUTION]: Several scripts, including scripts/story/generate-images.py and scripts/story/verify-images.py, use the sips utility via subprocess.check_output to validate the aspect ratio of generated images. This is a benign use of a native macOS system utility for media verification.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests textbook content (markdown) to generate slides and image prompts. The risk is limited as the skill uses structural delimiters (like HTML <details> blocks for prompts) and targets high-quality AI models, aligning with the low-severity risk typically associated with standard document processing workflows.
[SAFE]: The skill includes a non-portable hardcoded local path (/Users/dan/Documents/ws/...) in its documentation. While this is a best-practice violation for portability, it does not pose a security threat.

book-media-generator