The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill relies heavily on shell commands including curl, grep, sed, and jq to perform site audits. While these are standard tools, they execute with the agent's shell privileges.
[EXTERNAL_DOWNLOADS] (LOW): The skill uses npx lighthouse and npx puppeteer to perform performance and rendering audits.
Evidence: audit-workflow.md and technical-seo.md both contain npx commands.
Trusted Source: lighthouse and puppeteer are maintained by Google/Chrome, qualifying as trusted under [TRUST-SCOPE-RULE], which keeps this finding at LOW severity.
[PROMPT_INJECTION] (MEDIUM): This finding specifically addresses Category 8: Indirect Prompt Injection.
Ingestion points: audit-workflow.md (multiple locations, e.g., lines 27, 43, 85) uses curl to fetch content from user-provided or external URLs.
Boundary markers: Absent. Content is fetched and piped directly to text processing tools without delimiters or instructions to ignore embedded commands.
Capability inventory: The agent can execute shell commands, run Node.js tools, and write files to the local system (--output-path=./report.json).
Sanitization: None. The skill does not sanitize or escape the HTML/metadata content fetched from external sites.
Risk: A malicious website being audited could contain instructions (e.g., in HTML comments or meta tags) designed to hijack the agent's logic during the audit process.

seo-mastery