The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of multiple external dependencies from remote sources. This includes the nanobanana extension from a third-party GitHub repository (https://github.com/gemini-cli-extensions/nanobanana) and several Python libraries from the official package registry (edge-tts, kokoro, elevenlabs). Additionally, it performs a download of a 350MB text-to-speech model from HuggingFace when initializing the Kokoro engine.
[REMOTE_CODE_EXECUTION]: By requiring the installation and use of the nanobanana Gemini CLI extension and various Python packages, the skill introduces code from third-party sources into the local environment. These tools are invoked to handle core functionality such as image generation and editing.
[COMMAND_EXECUTION]: The skill makes extensive use of the run_terminal tool to execute shell commands for document conversion tasks using LibreOffice (soffice), image generation from PDFs (pdftoppm), and video assembly via ffmpeg. It also runs bundled Python scripts located in the scripts/ directory to manage text-to-speech workflows.
[PROMPT_INJECTION]: The skill processes research papers and user-provided outlines to draft presentation scripts, which establishes an indirect prompt injection surface. Malicious instructions embedded within a processed paper could potentially influence the agent's behavior during the drafting phase.
Ingestion points: Reads content from research papers and slide plans provided in the project workspace.
Boundary markers: The instructions do not define specific delimiters or warnings to separate external paper content from the agent's execution logic.
Capability inventory: The skill has access to terminal execution (run_terminal) and file system operations (read_file, write_file).
Sanitization: There is no mention of sanitization or verification processes for the content ingested from the external research papers.

making-academic-presentations