The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external websites via Lighthouse audits and uses that information to automatically implement code fixes. This creates a surface for indirect prompt injection, where an attacker-controlled website could craft metadata or page content designed to influence the agent's actions when it processes the audit report.
Ingestion points: scripts/audit.js and scripts/audit-api.js ingest content from arbitrary external URLs during the auditing process.
Boundary markers: The audit results are returned as structured JSON, but strings extracted from the audited sites (e.g., page titles, meta descriptions) are passed into the agent's context without specific delimiters or warnings to ignore embedded instructions.
Capability inventory: The agent is granted the capability to "automatically implement fixes" in the local codebase (file-write access) based on the audit findings.
Sanitization: The audit scripts do not sanitize or filter the content retrieved from the audited websites before presenting it to the agent.
[EXTERNAL_DOWNLOADS]: The scripts/audit-api.js script fetches audit data from the Google PageSpeed Insights API (googleapis.com). This is a well-known and trusted service for performance analysis.
[COMMAND_EXECUTION]: The skill launches a local instance of the Chrome browser to perform audits. The scripts/audit.js file uses the --no-sandbox flag when launching Chrome, which is a common configuration for headless environments but reduces the browser's internal security isolation.

roier-seo