The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from potentially untrusted external sources, which represents a surface for indirect prompt injection attacks.
Ingestion points: According to SKILL.md and references/skill-contract.md, the skill ingests data from scraped page text, customer reviews, comments, emails, and third-party data exports.
Boundary markers: The skill instructions explicitly direct the agent to treat these sources as untrusted evidence and state that it must not follow any embedded instructions found within the data.
Capability inventory: references/connectors.md defines integrations with various marketing execution tools, including Google Ads, HubSpot, Salesforce, and email providers like Resend and Mailchimp, which could be targets of malicious payloads if instructions were followed.
Sanitization: The skill requires the agent to label evidence (e.g., as 'measured' or 'user-provided') and distinguish direct observations from assumptions, providing a framework for safer data interpretation.
[NO_CODE]: The skill consists entirely of instructional markdown and configuration files with no executable scripts or binary dependencies provided.

consent-registry