The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates (assets/invoice_template.py and assets/report_template.py) that process untrusted data into PDF content. ReportLab Paragraph objects support XML-like inline markup tags (e.g., <link>, <b>, <font>). If an agent uses these templates to process external data containing malicious markup without sanitization, it could lead to the generation of deceptive or malicious PDF content.
Ingestion points: Data fields such as items and notes in create_invoice (assets/invoice_template.py) and sections or content in create_report (assets/report_template.py).
Boundary markers: Absent; data is directly interpolated into Paragraph strings.
Capability inventory: The skill uses SimpleDocTemplate.build() and canvas.save() to write generated PDF files to the local file system.
Sanitization: The provided example scripts do not demonstrate escaping or sanitization of input strings before they are passed to the ReportLab layout engine.

reportlab