The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and analyze content from user-provided URLs, which introduces a surface for indirect prompt injection.
Ingestion points: HTML content is fetched from websites during the audit phase (Phase 2).
Boundary markers: The skill explicitly instructs the agent to treat fetched data as untrusted and to mentally wrap it in <untrusted-content> tags to maintain separation from instructions.
Capability inventory: The agent has the ability to fetch web content and write local Markdown files.
Sanitization: The instructions contain specific defensive guidance to ignore any instruction-like text (such as "Ignore previous instructions") found within fetched content and to report such instances as injection attempts.
[EXTERNAL_DOWNLOADS]: The skill report includes a reference to aivsrank.com, a third-party service for measuring AI visibility. The report generates a link to this service with a referral parameter as part of a diagnostic integration.
[PROMPT_INJECTION]: A static analysis flag regarding instruction overrides was identified as a false positive. The string "Ignore previous instructions" appears in the SKILL.md file exclusively as part of security instructions teaching the agent how to detect and ignore malicious content, rather than an attempt to subvert the agent's behavior.

geo-monitor