The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes a vulnerability surface for indirect prompt injection or data poisoning by processing untrusted input into email templates without sanitization.
Ingestion points: The send_notification_email function in references/pythonflask-with-smtp.md and the sendEmail function in references/nodejs-with-sendgrid.md accept parameter data that is expected to be sourced from external user requests or database records.
Boundary markers: There are no boundary markers or delimiters specified in the templates to differentiate between structural content and user-provided data.
Capability inventory: The skill provides a significant capability for network communication (sending emails via mail.send and sgMail.send) and template rendering (render_template_string).
Sanitization: Sanitization is neglected in references/pythonflask-with-smtp.md where the |safe filter is applied to the content variable, instructing the template engine to bypass auto-escaping. This allows raw HTML or template directives to be processed at runtime.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic rendering of content based on runtime inputs.
Evidence: The use of render_template_string in Flask (references/pythonflask-with-smtp.md) and mjml_to_html for MJML compilation (references/email-templates-with-mjml.md).
Risk: These functions are used to assemble and process executable-like content from variable data, which can lead to injection attacks if the data sources are untrusted and not validated before interpolation.

email-service-integration