The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites through its audit script (scripts/wcag22_audit.cjs), which scrapes text elements like headings, link labels, and page titles. This content is then included in reports analyzed by the agent, creating a potential vector where instructions embedded in a scanned page could influence the agent.
Ingestion points: The audit script extracts heading.textContent, link.textContent, document.title, and image.alt from URLs provided during the audit.
Boundary markers: The skill lacks explicit instructions for the agent to treat scraped data as untrusted or to ignore embedded instructions.
Capability inventory: The skill environment allows for shell command execution (node, npm) and network access via a browser automation tool (Playwright).
Sanitization: While the script performs basic trimming and length-limiting on scraped strings, it does not sanitize the content for potential prompt injection markers.
[COMMAND_EXECUTION]: The workflow relies on the agent executing a bundled Node.js script (scripts/wcag22_audit.cjs). This script performs filesystem operations (writing reports) and network requests (navigating to URLs).
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the playwright and axe-core Node.js packages from the public NPM registry. These are reputable, well-known libraries used for their intended purpose in this context.
[DYNAMIC_EXECUTION]: The audit script uses page.evaluate() to run analysis logic within the browser context. This is a functional requirement for auditing the DOM of rendered pages.

typo3-accessibility