The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by crawling sitemaps and scanning live HTML from user-provided URLs. This presents an attack surface where malicious content on the target site could attempt to influence the agent's behavior.
Ingestion points: The discover-urls.mjs script fetches and parses external XML sitemaps, and accessibility.spec.js uses Playwright to navigate to and scan live URLs.
Boundary markers: The skill explicitly instructs the agent that automated results are only evidence and that a human review (HITL) is mandatory for criteria like alt-text quality and legal conformance.
Capability inventory: The skill can write audit reports, generate accessibility statements, and perform template linting on the local filesystem.
Sanitization: Uses fast-xml-parser for sitemaps and runs browser-based scans within a Playwright instance, providing some isolation.
[EXTERNAL_DOWNLOADS]: The skill's configuration installs standard, well-known development dependencies and browser binaries.
The package.json file includes standard accessibility and testing libraries such as @axe-core/playwright and @playwright/test (from Microsoft).
The Skills.sh script facilitates local setup without downloading arbitrary remote payloads.
[COMMAND_EXECUTION]: The skill involves executing shell commands to run tests and generate reports (e.g., npm run a11y:all). These commands are standard for a development/auditing workflow and are scoped to the project directory.

typo3-wcag22-aa-agentic