The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of email templates that ingest and render untrusted user data (e.g., name, orderNumber). This represents a vulnerability surface where malicious input could attempt to influence the agent's output or perform cross-site scripting (XSS) if the resulting HTML is rendered in an insecure context.
Ingestion points: Component props in SKILL.md, references/I18N.md, and references/PATTERNS.md (e.g., WelcomeEmailProps, OrderConfirmationProps, NotificationProps).
Boundary markers: Absent. The instructions do not specify using delimiters or warnings to prevent the agent from obeying instructions embedded within the template data strings at runtime.
Capability inventory: The skill uses the render function from react-email to generate HTML/Text and provides examples for sending via various email providers' APIs (resend, nodemailer, mailgun.js, sgMail).
Sanitization: Absent. While React's JSX provides default escaping for HTML, the skill does not explicitly instruct the agent to sanitize inputs for plain-text rendering or to validate the structure of ingested data to prevent data-driven attacks.
[METADATA_POISONING]: A discrepancy exists between the provided author context ("blockmatic") and the manifest in SKILL.md, which lists "Resend" as the author. While all provided links are to legitimate resend.com and react.email resources, such inconsistencies in reported provenance can be deceptive and may hinder accurate supply-chain auditing.

react-email