The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text and document summaries which are then sent to an external API for slide generation.
Ingestion points: Content is passed to generate_slides.py, create_pdf_slides.py, and generate_narration.py through the --content or user_input parameters.
Boundary markers: The instructions guide the agent to structure content (e.g., using titles and sections). The Python scripts use JSON encoding for API payloads, providing structural separation.
Capability inventory: The skill has network access (requests) and file writing capabilities (download_slides_pages_voices.py).
Sanitization: The scripts rely on standard JSON serialization for data transmission to the 2slides service.
[EXTERNAL_DOWNLOADS]: The download_slides_pages_voices.py script facilitates the download of ZIP archives containing presentation assets.
Evidence: The script includes a download_pinned_https function that implements several defensive measures: it requires HTTPS, validates that hostnames resolve to global (public) IP addresses, enforces a 128 MiB size limit, and prevents path traversal via the safe_user_path utility.
[COMMAND_EXECUTION]: The skill operates by invoking various Python scripts to interact with the 2slides API.
Evidence: Functionality is implemented via scripts like generate_slides.py and get_job_status.py. These scripts use argparse for parameter handling and include regex validation for identifiers like job_id to mitigate potential injection risks.

2slides-ppt-generator