The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and executes Python scripts at runtime to handle PowerPoint (.pptx) file extraction and export. This involves writing logic to a file and executing it based on user-provided inputs.
[COMMAND_EXECUTION]: Employs browser automation (browser-use) to execute arbitrary JavaScript (eval) on external websites to programmatically extract CSS styles and brand assets. This executes code in a context influenced by external, potentially untrusted web content.
[EXTERNAL_DOWNLOADS]: Instructs the agent to install external software dependencies (python-pptx, Pillow) via pip during the conversion process.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted data from multiple external vectors including PowerPoint files and third-party websites. It lacks explicit sanitization or strict boundary markers when interpolating this data into the final presentation output.
Ingestion points: User-provided .pptx files and external website content extracted via browser automation.
Boundary markers: No specific delimiters or "ignore instructions" warnings are utilized during content interpolation.
Capability inventory: File system access (.octave-decks/), Python subprocess execution, and browser-based code execution.
Sanitization: Content is extracted and mapped directly to slide templates without validation for malicious instructions.

deck