The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data by interpolating it into email bodies, which presents a surface for potential misuse if the agent processes malicious external content.
Ingestion points: The username and orderReference parameters in the sendOrderConfirmationEmail function example within SKILL.md.
Boundary markers: The implementation example does not demonstrate the use of delimiters or warnings to isolate untrusted data within the email body.
Capability inventory: The skill has the capability to perform network outcalls via EmailClient.sendServiceEmail to an external gateway, as shown in SKILL.md.
Sanitization: The documentation and code examples do not include explicit sanitization or escaping of the interpolated string components.
[SAFE]: The skill follows secure practices by using platform-injected secrets (INTEGRATIONS_GATEWAY_API_KEY) rather than hardcoding credentials, and it relies on versioned dependencies from the official vendor namespace.

extension-email