The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates the risk of indirect prompt injection by explicitly instructing the agent to treat all fetched page content as untrusted data. It provides clear directives to analyze content for SEO purposes but to never follow instructions embedded in HTML, meta tags, or page copy.
Ingestion points: web_fetch and curl operations for website analysis, and DOM queries using document.querySelectorAll in the browser tool.
Boundary markers: Explicit system-level warning in SKILL.md to treat fetched data as a prompt-injection surface.
Capability inventory: Limited to web retrieval and analysis; no evidence of arbitrary file-writing, system commands, or unsafe code execution.
Sanitization: Clear instruction to separate content analysis from instruction following.
[DATA_EXFILTRATION]: The skill references established industry tools (e.g., Google Search Console, Ahrefs, Semrush, DataforSEO) and official documentation from trusted technology providers like Google, Microsoft, and Cloudflare. There are no hardcoded credentials, API keys, or attempts to access sensitive local system files like .ssh or .aws credentials.
[COMMAND_EXECUTION]: Shell-related tools such as curl are used strictly for retrieving public metadata from websites as part of the SEO audit workflow. There is no usage of sudo, chmod, or other privilege escalation techniques.
[OBFUSCATION]: A thorough manual and automated review of the markdown and reference files found no evidence of Base64-encoded commands, zero-width characters, homoglyph attacks, or steganographic patterns hiding malicious URLs or instructions.

seo-audit