The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external extension for the Gemini CLI and downloads a machine learning model at runtime.
Extension: https://github.com/gemini-cli-extensions/nanobanana is installed via gemini extensions install.
Model: hexgrad/Kokoro-82M is downloaded from HuggingFace when scripts/batch_tts_kokoro.py is run.
[COMMAND_EXECUTION]: The skill relies on executing various system-level commands to perform its tasks.
Terminal Tools: Uses ffmpeg for video assembly, soffice (LibreOffice) for PDF conversion, and pdftoppm for image extraction.
Shell Usage: Includes bash commands for file manipulation and conversion in the reference guides.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it processes untrusted external data.
Ingestion points: Research papers and user-provided outlines (defined in SKILL.md Stage 1).
Boundary markers: Lacks explicit delimiters or instructions to ignore embedded commands within the ingested research papers.
Capability inventory: Terminal execution access via run_terminal and the ability to write files (write_file).
Sanitization: No sanitization of the research paper text or user outline is performed before processing.

making-academic-presentations