The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to perform audits on external websites, which involves reading and analyzing untrusted data such as meta tags, robots.txt files, and page content. This behavior establishes an attack surface where a malicious site could attempt to influence the agent's actions through embedded instructions.
Ingestion points: The audit process outlined in SKILL.md requires fetching data from various external URLs and site components.
Boundary markers: The instructions do not include specific delimiters or safety warnings to prevent the agent from following instructions discovered during the audit process.
Capability inventory: Although the skill consists only of documentation and instructions, an agent utilizing it would typically have capabilities for web browsing and potentially file system access to generate reports.
Sanitization: The skill lacks protocols for sanitizing external inputs or distinguishing between site content and agent instructions.

seo-audit