The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts doc_to_md.py and project_manager.py use subprocess.run() to invoke external system utilities like pandoc and node, as well as other internal Python scripts. While arguments are passed as a list (reducing shell injection risk), this execution surface relies on the security of the underlying system binaries and the integrity of input filenames.
[EXTERNAL_DOWNLOADS]: web_to_md.py and web_to_md.cjs fetch content from arbitrary user-provided URLs and download images. The Python implementation explicitly disables SSL certificate verification using verify=False, which increases exposure to Man-in-the-Middle (MitM) attacks during content ingestion.
[PROMPT_INJECTION]: The skill has a significant surface for indirect prompt injection, as it ingests untrusted text from external documents and URLs to drive the presentation strategy and slide generation.
Ingestion points: Source materials processed in scripts/pdf_to_md.py, scripts/doc_to_md.py, and scripts/web_to_md.py are converted to Markdown and read by the agent.
Boundary markers: Absent. The skill does not wrap the processed text in delimiters or provide instructions for the agent to ignore commands embedded in the source material.
Capability inventory: The agent has capabilities to write files, perform network downloads, and execute system commands via the provided toolset.
Sanitization: While filenames are sanitized for filesystem safety, the text content is not filtered for instruction-like patterns.

ppt-master