The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/local_artifact_smoke.py script executes a generated Node.js script via subprocess.run to leverage Playwright for rendering HTML content into PDF and PNG images. The directory paths within the script are sanitized using JSON encoding to prevent injection.
[EXTERNAL_DOWNLOADS]: The skill depends on several external libraries and tools, specifically python-pptx for generating PowerPoint files and the notebooklm-py utility for advanced content synthesis. These dependencies are documented and required for full functionality.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted articles and URLs while possessing the ability to write files and execute local rendering commands.
Ingestion points: Input data sources include Markdown files from a vault and content retrieved from external URLs as described in SKILL.md and references/providers.md.
Boundary markers: Multiple prompt templates (e.g., references/prompt-ppt.md) explicitly instruct the agent to remain faithful to the source content and avoid introducing external facts.
Capability inventory: The skill can write files to the local file system and execute shell commands via Node.js for rendering purposes.
Sanitization: Content used in HTML generation is sanitized with html.escape in the scripts/local_artifact_smoke.py utility.

soia-pkm-transform-article-slides