The Agent Skills Directory

Indirect Prompt Injection (LOW): The scripts audit.js and audit-api.js ingest untrusted data from external websites and APIs which could contain malicious instructions designed to influence the agent's subsequent behavior.
Ingestion points: audit.js (extracts titles, descriptions, and element snippets from audited URLs), audit-api.js (ingests data from the PageSpeed Insights API).
Boundary markers: Absent. The output is printed as JSON or raw text to stdout without clear delimiters or 'ignore' instructions for the agent.
Capability inventory: audit.js has the capability to write files to the local disk via fs.writeFileSync.
Sanitization: Absent. The scripts pass through external content like audit descriptions and element snippets directly into the output.
Data Exposure & Exfiltration (LOW): The audit.js script allows arbitrary file writing which could be abused if an attacker influences the command-line arguments.
Evidence: The --save flag in audit.js uses path.resolve(save) and fs.writeFileSync without restricting the output directory, potentially allowing the overwriting of sensitive files if the agent is misdirected.
Network Operations (LOW): audit-api.js performs network requests to googleapis.com.
Evidence: Uses fetch to call the PageSpeed Insights API. While targeting a reputable Google domain, the domain is not on the predefined internal whitelist.

roier-seo