The Agent Skills Directory

[PROMPT_INJECTION]: The automated detector flagged a high-severity alert for the phrase 'ignore previous instructions' in SKILL.md and audit-workflow.md. Analysis confirms this is a benign false positive. The text is part of a dedicated security section instructing the agent on how to securely handle external content, specifically telling it to disregard adversarial statements like 'ignore previous instructions' found on external sites.
[INDIRECT_PROMPT_INJECTION]: The skill inherently interfaces with untrusted data by instructing the agent to perform SEO audits on external websites (ingesting HTML, robots.txt, and sitemaps). The skill exhibits excellent defense-in-depth by providing explicit security boundaries, instructing the agent to treat all external inputs as data rather than commands, wrap inputs in explicit XML boundary markers (<untrusted_fetched_content>), and report injection attempts verbatim instead of executing them.
[EXTERNAL_DOWNLOADS]: The skill mentions external tools and frameworks such as Lighthouse, Astro integrations, and Next.js/Nuxt packages. All references point to standard, well-known enterprise open-source packages and official Google/Cloudflare documentation, presenting no supply chain or unverified dependency risks.

seo-mastery