The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and analysis of content from external URLs, which creates a potential surface for indirect prompt injection if target pages contain malicious instructions intended to influence the agent's behavior.
Ingestion points: The AccessibilityAuditor and other testing classes in resources/implementation-playbook.md utilize Puppeteer to navigate to and process content from external URLs provided at runtime.
Boundary markers: There are no explicit delimiters or specific instructions for the agent to treat ingested web content as data only and ignore any potential embedded instructions.
Capability inventory: The skill utilizes browser automation (Puppeteer) and provides templates for CI/CD integration, which could be leveraged if an agent were to follow malicious instructions found on a scanned page.
Sanitization: The skill includes a utility function, escapeReportText, designed to sanitize findings before they are rendered into reports, which helps mitigate potential cross-site scripting risks in the final output.
[EXTERNAL_DOWNLOADS]: The skill references and provides implementation guidance for several well-known industry packages and official automation tools.
The documentation recommends the use of established packages such as @axe-core/puppeteer, puppeteer, jest-axe, and pa11y for automated accessibility testing.
The provided CI/CD integration examples utilize official GitHub Actions, including actions/checkout and actions/setup-node.

accessibility-compliance-accessibility-audit