The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from external sources, which could potentially contain malicious instructions targeting the AI agent.
Ingestion points: Incoming email content, including Subject, TextBody, and HtmlBody fields, is delivered via webhooks to be processed by the agent as described in postmark-inbound/SKILL.md and postmark-inbound/references/payload-structure.md.
Boundary markers: There are no explicit instructions or delimiters provided within the skill to designate incoming email content as untrusted or to warn the agent against executing instructions contained within that data.
Capability inventory: The skill provides documentation and examples for high-privilege operations including file system writes (saving attachments in postmark-inbound/references/handler-examples.md), database interactions, and network requests (sending reply emails).
Sanitization: While the documentation mentions that Handlebars templates escape content by default (postmark-templates/references/handlebars-syntax.md), it does not provide specific guidance or implementation details for sanitizing incoming email bodies to prevent prompt injection attacks.

postmark