The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection via the generated HTML viewer. The Python script scripts/generate_viewer.py processes JSON data (which can be sourced from untrusted user inputs or other skills) and interpolates it directly into HTML templates without sanitization. This allows malicious HTML or JavaScript embedded in the title or content fields of the input JSON to be executed when a user opens the resulting HTML file.
Ingestion points: The load_ppt_data function in scripts/generate_viewer.py reads data from ppt_data.json.
Boundary markers: None identified; the script does not include instructions to the agent or delimiters to distinguish between content and control logic in the processed data.
Capability inventory: The skill performs file system operations (writing HTML, creating directories, and copying CSS/JS assets) and prepares materials for video composition.
Sanitization: Absent. The generate_viewer_html function uses f-strings and template.replace to insert slide titles and content directly into the HTML body without escaping special characters.

nanobanana-ppt-visualizer